CN111695426B - Behavior pattern analysis method and system based on Internet of things - Google Patents

Behavior pattern analysis method and system based on Internet of things Download PDF

Info

Publication number
CN111695426B
CN111695426B CN202010380711.1A CN202010380711A CN111695426B CN 111695426 B CN111695426 B CN 111695426B CN 202010380711 A CN202010380711 A CN 202010380711A CN 111695426 B CN111695426 B CN 111695426B
Authority
CN
China
Prior art keywords
user
sequence
activity
sensor
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010380711.1A
Other languages
Chinese (zh)
Other versions
CN111695426A (en
Inventor
靳浩
王厅玮
张庆
赵成林
彭木根
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN202010380711.1A priority Critical patent/CN111695426B/en
Publication of CN111695426A publication Critical patent/CN111695426A/en
Application granted granted Critical
Publication of CN111695426B publication Critical patent/CN111695426B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Data Mining & Analysis (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Biomedical Technology (AREA)
  • Primary Health Care (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • Signal Processing (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)

Abstract

The invention provides a behavior pattern analysis method and system based on the Internet of things, wherein the method comprises the following steps: obtaining time sequence data of a user to be detected, wherein the time sequence data comprises the following steps: the user to be tested executes at least one target time point of the preset activities according to the appointed steps, and sensor events corresponding to the target time point describe the state change condition of each sensor at the target time point; determining a position sequence of a user to be detected according to the time sequence data; determining a user position sub-sequence based on matrix representation according to the time sequence data and the position sequence; and determining an activity pattern diagram of the user to be tested according to the user position subsequence based on matrix representation, and giving out a behavior pattern analysis result of the user to be tested from the aspects of qualitative and quantitative based on the activity pattern diagram. The invention can be used for classifying and evaluating the dementia activity modes of users, realizes automatic segmentation of the sensor time sequence, and improves the usability and the interpretability of data analysis based on the sensor time sequence.

Description

Behavior pattern analysis method and system based on Internet of things
Technical Field
The invention relates to the technical field of data analysis, in particular to a behavior pattern analysis method and system based on the Internet of things.
Background
With the rapid development of the internet of things, pattern analysis based on intelligent equipment and sensor data is mainly applied to the smart city field (such as traffic pattern analysis, temperature and climate analysis and the like), the smart home field (such as dementia monitoring, home safety monitoring and the like) and the smart wearable equipment field (such as health detection, movement monitoring and the like), and the method also provides possibility for supporting low-cost illness state monitoring based on embedded living environments.
In a condition monitoring system based on environmental sensor data, by deploying sensors and monitoring devices in a home environment, when a user performs a specified activity comprising several steps within the environment, the user's behavior triggers a change in the state of the environmental sensor, and the monitoring system can obtain the user's behavior pattern by collecting a data sequence generated from the sensor's state and analyzing the data sequence. Based on the analysis results of the behavior patterns, intelligent assistance can be provided for the life of the user, for example, behavior assistance prompts of the user can be given, alarm emergency is carried out when necessary, and dementia condition analysis is carried out based on the behavior patterns of the user so as to find out early dementia patients.
Compared with the solutions based on audio and video data, wearable sensors and multi-mode sensors, the solution based on the environment sensor has the characteristics of low economic cost, remarkably lower data volume and calculation complexity than the data volume based on audio and video data analysis, data acquisition only when a user is in the detection coverage range, low invasiveness to the privacy of the user and the like.
There are some environmental sensor-based activity pattern analysis and disease monitoring research efforts. From the feature type, the dementia evaluation methods based on the environmental sensor are mainly classified into two categories, namely, an activity feature-based research result and a walking pattern-based research result. When the existing research results are used for analyzing the activity mode, the activity mode is modeled based on different characteristics, and the main problems include: first, existing schemes typically rely on a large number of manual operations, including data tagging and pre-segmentation, which are difficult to put into practical use due to the high cost of manual operations; second, pattern classification based on traditional machine learning algorithms lacks interpretability; third, the granularity of the labels and pre-segments is coarse, and no solution for user activity pattern learning and representation for fine-grained activity has been found.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems in the related art to some extent.
Therefore, an object of the present invention is to provide a behavior pattern analysis method based on the internet of things, which can realize automatic segmentation of sequences, perform behavior pattern analysis from a qualitative and quantitative perspective based on fine-granularity activity patterns, and can be used for classification and evaluation of dementia behavior patterns.
Another object of the present invention is to provide a behavior pattern analysis method system based on the internet of things.
In order to achieve the above object, a behavior pattern analysis method based on the internet of things according to an embodiment of the first aspect of the present invention includes: obtaining time sequence data of a user to be tested, wherein the time sequence data comprises the following steps: the user to be tested executes at least one target time point of preset activities according to the appointed steps, and sensor events corresponding to the target time point describe the state change condition of each sensor at the target time point; determining a position sequence of the user to be detected according to the time sequence data; combining the position sequences according to the time sequence data, and determining a user position subsequence based on matrix representation; and determining an activity pattern diagram of the user to be tested according to the user position subsequence based on the matrix representation, and analyzing a behavior pattern of the user to be tested based on the activity pattern diagram.
According to the behavior pattern analysis method based on the Internet of things, based on the home environment where the environment sensor is deployed, the user performs activities in the environment according to the predefined steps, the state change of the environment sensor is triggered, the corresponding sensor state sequence is obtained, sensor time sequence data of the user is formed, based on the data, automatic segmentation of the sequence is achieved by adopting graph wavelet transformation, the cost of manually processing the data is reduced, fine-granularity user activity pattern recognition is adopted, the usability and the interpretability of data analysis based on the sensor time sequence are improved, and the privacy invasion of the user and discomfort caused by monitoring based on a wearable sensor are reduced.
In order to achieve the above object, an behavioral pattern analysis system based on the internet of things according to an embodiment of the second aspect of the present invention includes: the acquisition module is used for acquiring time sequence data of the user to be detected, wherein the time sequence data comprises the following components: the user to be tested executes at least one target time point of preset activities according to the appointed steps, and sensor events corresponding to the target time point describe the state change condition of each sensor at the target time point; the data denoising function module is used for determining the position sequence of the user to be detected according to the time sequence data; a representation learning function module for determining a user location sub-sequence based on a matrix representation in accordance with the time-series data in combination with the location sequence; and the activity mode construction function module is used for determining an activity mode diagram of the user to be tested according to the user position subsequence based on the matrix representation and analyzing the behavior mode of the user to be tested based on the activity mode diagram.
According to the behavior pattern analysis system based on the Internet of things, provided by the embodiment of the invention, based on the home environment where the environment sensor is deployed, the user performs activities in the environment according to predefined steps, triggers the state change of the environment sensor, acquires the corresponding sensor state sequence, forms the sensor time sequence data of the user, realizes automatic segmentation of the sequence by adopting graph wavelet transformation based on the data, reduces the cost of manually processing the data, adopts fine-granularity user activity pattern recognition, improves the usability and the interpretability of data analysis based on the sensor time sequence, and reduces the privacy invasion of the user and discomfort caused by monitoring based on the wearable sensor.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
fig. 1 is a flowchart of a behavior pattern analysis method based on the internet of things according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a typical reference system scenario and corresponding sensor deployment results presented by an embodiment of the present invention;
FIG. 3 is a flow chart of data denoising in a template construction task according to an embodiment of the present invention;
FIG. 4 is a flow chart of data denoising in dementia classification and assessment tasks as proposed by an embodiment of the present invention;
FIG. 5 is a flow chart of sequence segmentation proposed by an embodiment of the present invention;
FIG. 6 is a flow chart of a representation of a sequence proposed by an embodiment of the present invention;
FIG. 7 is a flow chart of exemplary movement pattern mining in performing template building tasks in accordance with an embodiment of the present invention;
FIG. 8 is a flow chart of exemplary movement pattern mining in performing dementia assessment tasks in accordance with an embodiment of the present invention;
FIG. 9 is a flow chart of an activity pattern build when performing a template build task in accordance with an embodiment of the present invention;
FIG. 10 is a flow chart of the activity pattern construction in performing dementia assessment tasks according to an embodiment of the present invention;
FIG. 11 is a flow chart of dementia assessment in performing template building tasks in accordance with an embodiment of the present invention;
FIG. 12 is a flow chart of dementia assessment in performing dementia assessment tasks according to an embodiment of the present invention;
FIG. 13 is a flow chart of an exemplary template building task presented by an embodiment of the present invention;
FIG. 14 is a flow chart of dementia classification and assessment tasks as proposed by an embodiment of the present invention;
fig. 15 is a block schematic diagram of an behavioral pattern analysis system based on the internet of things according to an embodiment of the present invention;
fig. 16 is a block diagram of an behavioral pattern analysis system based on the internet of things according to an embodiment of the invention.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the invention. On the contrary, the embodiments of the invention include all alternatives, modifications and equivalents as may be included within the spirit and scope of the appended claims.
The present application is made based on the knowledge and findings of the inventors of the following problems:
advantages of environmental sensor data based condition monitoring systems are mainly manifested in the following aspects: firstly, the system has low energy consumption and low equipment economic cost, and usually does not need professional maintenance, the environment sensor can be deployed in the daily home environment of the user or a community medical service station, the user does not need to go to a hospital, and can accept monitoring only in the daily home environment, so that the traffic and time cost of the user for going to the hospital for inspection can be saved, and the medical cost of the user can be saved; secondly, in the monitoring system based on the environmental sensor data, compared with a monitoring system of a wearable sensor, a user does not need to wear the sensor, and meanwhile, the privacy invasiveness of the data collected based on the environmental sensor to the user is lower, so that the monitoring system has better user experience; third, compared with a data acquisition method based on video equipment such as a camera, the data amount is small. Therefore, the condition monitoring based on the environmental sensor data has wide application prospect.
In the condition monitoring research based on the environmental sensor data, the analysis of the user behavior pattern based on the environmental sensor data is a key technology for realizing condition monitoring and is also a technical difficulty.
The invention is deeply studied for a method for dementia evaluation based on a time sequence of an environmental sensor, and an effective method for dementia behavior pattern analysis based on the time sequence of the sensor is not found at present, and main challenges comprise the following aspects: firstly, in a monitoring system based on environmental sensor data, due to dense deployment and delayed closing of sensors, noise is contained in the collected sensor data, so that analysis effect based on the sensor data is reduced, and an effective sensor data denoising method is lacking at present; secondly, most of analysis of sensor data in the existing system is based on manual pre-segmentation operation, so that high data analysis cost is caused, and an effective method for segmenting based on unsupervised learning and oriented to a sensor time sequence is lacking at present; third, in the prior art, activity statistics behavior features based on manual construction are generally adopted, and a solution for learning a fine-grained user behavior pattern is lacking.
Based on the above reasons, the embodiment of the invention provides a behavior pattern analysis method and a behavior pattern analysis system based on the Internet of things, and the behavior pattern analysis method and the behavior pattern analysis system based on the sensor can realize automatic segmentation of the time sequence of the sensor on the basis of greatly reducing the data volume acquired by the behavior pattern analysis method of the Internet of things, reduce the cost of manually processing the data, improve the usability of data analysis based on the time sequence of the sensor, and reduce the privacy invasion of users and discomfort caused by monitoring based on a wearable sensor. Meanwhile, the invention is based on the user activity pattern analysis of fine granularity, can evaluate dementia from the qualitative and quantitative angles, and improves the interpretation of analysis results.
The method and system for analyzing the behavior pattern based on the internet of things according to the embodiment of the invention are described below with reference to the accompanying drawings, and the method for analyzing the behavior pattern based on the internet of things according to the embodiment of the invention will be described first with reference to the accompanying drawings.
Fig. 1 is a flowchart of a behavior pattern analysis method based on the internet of things according to an embodiment of the present invention.
As shown in fig. 1, the behavior pattern analysis method based on the internet of things comprises the following steps:
In step S101, time-series data of a user to be tested is acquired, the time-series data including: and the user to be tested executes at least one target time point of the preset activities according to the appointed steps, and sensor events corresponding to the target time point describe the state change condition of each sensor at the target time point.
When the method is applied specifically, the steps of daily activities and corresponding activities are formulated in daily home scenes where the activity detection sensor and the auxiliary sensor are deployed. The user completes corresponding daily activities based on established typical activity steps, when the user performs activities according to predefined steps in the environment, the environment sensor is triggered to cause state change of the environment sensor, the computing equipment deployed in a daily home scene acquires corresponding sensor time-series state data, based on the data, activity mode representation learning of the sensor time-series data of the user is completed, and based on user classification and illness state evaluation representing learning results.
The embodiment of the invention will be described by taking a scenario shown in fig. 2 as an example, wherein as shown in fig. 2, related devices mainly comprise home devices, motion detection sensors, auxiliary sensors, networking devices and computing devices for completing user activity pattern representation learning and dementia classification evaluation. The main functions of the respective devices related to user activity pattern learning and classification evaluation are given below.
(1) Household equipment
The household equipment comprises corresponding household equipment which is arranged in a household environment required by a user to execute a predefined activity and mainly comprises electric appliances, furniture and articles, wherein a sensor is arranged on the household equipment, and conditions are provided for the user to finish the specified activity and the corresponding steps. The types of sensors deployed include, but are not limited to, motion detection sensors and auxiliary sensors.
(2) Motion detection sensor
The motion detection sensor refers to a passive infrared sensor attached to a ceiling, a wall, or a stationary object, and is in an "on" state when a participant moves within its detection range, and changes to an "off" state when the participant remains stationary or leaves its detection range. The motion detection sensor is mainly used for detecting whether a user is in the coverage range of the sensor, and positioning of the user can be completed according to the on-off state of the sensor, so that track information of the user is provided for user activity pattern analysis.
(3) Auxiliary sensor
Auxiliary sensors are those attached to the necessary items that the user needs to interact with to detect if the status of the items changes, e.g., if the door is open, if the faucet is open, etc. The auxiliary sensor is mainly used for providing logic information of user activities.
(4) Networking equipment
The networking equipment mainly refers to networking related equipment used for connecting electrical equipment, household equipment, sensors, articles and computing equipment for completing user activity mode learning and classification evaluation, and can be used for networking in a wired or wireless mode.
(5) Computing device for completing user activity pattern learning and classification evaluation
The device communicates with a sensor network through network equipment, collects sensor time sequence data information related to the sensors, and mainly comprises the steps of receiving and storing the sensor time sequence data, completing learning and classifying evaluation functions aiming at a user activity mode, and giving a learning result aiming at the user activity mode and a classifying evaluation result aiming at the user activity mode.
Based on the above sensors disposed on the respective devices/articles, when a user performs an activity in a home environment according to a predefined procedure, the motion detection sensor and the auxiliary sensor are triggered, and when the state of any one sensor is changed, a sensor event is established, where the sensor event includes the number of the changed sensor, the time when the state of the sensor is changed, the changing manner of the state of the sensor, and the states of all the sensors at the time (where the state of the changed sensor is based on the changed state). These sensor events are sent to the computing device over the sensor network, stored in chronological order. When the user stops the activity, all sensor events of the user are integrated into sensor time series data for the computing device to analyze by adopting the user activity pattern learning and dementia classification evaluation method based on the graph signals and the graph theory.
In step S102, a position sequence of the user to be measured is determined according to the time-series data.
In one embodiment of the present invention, determining a location sequence of a user to be measured according to time sequence data includes: and denoising the time sequence data, so as to determine the position sequence of the user to be detected.
It can be understood that the embodiment of the invention provides a data denoising method and a data denoising flow based on a graph signal processing theory, thereby solving the denoising problem of sensor data.
Further, in one embodiment of the invention, the categories of sensors include: the motion class carries out denoising processing on the time sequence data so as to determine the position sequence of the user to be detected, and the method comprises the following steps: according to time sequence data of a plurality of training users, using a motion sensor as a node, and using the number of times that every two motion sensors are in an on state in adjacent records as an edge connecting the two nodes to construct a graph; taking the sensor states corresponding to each target time point in the time sequence data of the user to be detected as graph signals, and performing graph wavelet transformation on the graph signals to obtain transformation results corresponding to each target time point; according to the transformation result, determining the identification of the target sensor corresponding to each target time point and taking the identification as the user position identification, wherein the actual position distance between the target sensor corresponding to the target time point and the user to be detected is the smallest at the target time point, and the category of the target sensor is a motion category; and forming a position sequence according to the user position identifiers corresponding to the target time points.
It can be understood that according to the movement characteristics of the user, the embodiment of the invention provides a method for representing the sensor time series data, the problem of segmentation of the sensor time series data is solved based on graph wavelet transformation by using the sequence obtained by representation, and the method for representing the sequence and the method for segmenting the sequence are provided, in particular: according to the states of the motion sensors of all target time points, an liveness sequence is constructed, and liveness sequence diagram signals are generated; processing the liveness sequence chart signal through chart wavelet transformation, and calculating the frequency category of each node; and carrying out segmentation processing on the position sequence based on the frequency class difference between the adjacent nodes.
Specifically, the embodiment of the invention provides a data denoising method based on a graph signal processing theory, and a data denoising functional module adopts the method to denoise a multidimensional sensor state sequence to obtain a motion detection sensor closest to a user in each record, and identifies the position of the user in the record based on the sensor. Specifically, the denoising process is modeled as identifying a core sensor problem in each record, the motion sensor state in each record is processed, the motion detection sensor closest to the tested person is taken as the core sensor of the record, and other motion detection sensors in the on state are false alarm sensors.
As shown in fig. 3, the data denoising process in the template construction task includes the following steps:
step 1: the data denoising function module inputs sensor sequence data.
Step 2: and taking the motion sensors as nodes, and taking the times that every two motion sensors are in an on state in adjacent records of training set data as edges for connecting the two nodes to construct a graph.
Step 3: the map signal is constructed based on the motion sensor states in the sensor sequence data.
Step 4: and determining the expansion scale search range.
Step 5: and selecting the wavelet expansion scale of the optimal graph for all records in the training set.
Step 6: judging the type of a sensor event, and if the sensor event is a record from closed to open, selecting the number of the open sensor as an actual core sensor of the record; if the sensor event is a record from on to off, step 7 is performed.
Step 7: and using a prediction core sensor obtained under the wavelet expansion scale of the optimal graph as a core sensor of the record.
Step 8: and identifying the position of the user by adopting each recorded core sensor, wherein the sequence formed by all the recorded core sensors is the position sequence of the user.
In addition, the data denoising process in the dementia classification and assessment task is shown in fig. 4, and steps 1, 2, 3, 7 and 8 in the data denoising process in the template construction task shown in fig. 3 are executed, so that redundancy is avoided and redundant description is omitted.
In step S103, a user position sub-sequence based on the matrix representation is determined from the time-series data in combination with the position sequence.
Wherein in one embodiment of the invention, determining a user location sub-sequence based on a matrix representation based on time series data in combination with a location sequence comprises: carrying out segmentation processing on the position sequence according to the time sequence data to obtain a user position sub-sequence; the user position sub-sequence is converted into a matrix representation based user position sub-sequence.
It can be understood that this step mainly completes the functions of segmenting the sensor sequence and representing the sequence, converting the user's position sequence and the sensor data sequence into a sub-sequence of user positions based on matrix representation, specifically the data segmentation functions are: segmenting a sensor data sequence generated when a user executes the whole activity and a position sequence generated by a denoising module when the user acts; the sequence representation functions are: for generating a matrix representation of the sequence, converting the segmented user position sub-sequence into a matrix-based position sub-sequence. Based on the step, the problems of learning and representing the fine granularity active mode based on the sensor time series data are solved, and meanwhile, the problems of classifying and evaluating dementia patients are solved based on the fine granularity active mode representation learning method of the sensor time series data.
The sequence segmentation and sequence representation functions will be described in detail below, respectively.
Further, in one embodiment of the present invention, the processing of the position sequence based on the time series data includes: constructing an activity sequence according to the state of the motion sensor of the time sequence data; taking all sensor records as nodes, and correspondingly taking adjacent records as equal weight edges to construct a sequence diagram based on the adjacent relation of the sensor sequences; according to the liveness sequence, constructing a liveness-based graph signal by combining sequence graphs of adjacent relations; according to the graph signals based on the liveness, calculating the frequency category of each node in the sequence graph of the adjacent relation; and carrying out segmentation processing on the position sequence based on the frequency class difference among the nodes connected by the edges.
Specifically, the embodiment of the invention provides a sequence segmentation method based on graph wavelet transformation, and a sequence segmentation submodule adopts the method to carry out sequence segmentation. The activity sequence is frequency classified based on a graph wavelet transform, and sequence segmentation is performed according to the frequency class, wherein the activity sequence is a representation sequence constructed based on the sensor data sequence. Specifically, as shown in fig. 5, the sequence segmentation includes the steps of:
Step 1: the data segmentation submodule receives the sensor sequence data and the user position sequence data generated by the data denoising module.
Step 2: based on the motion sensor state, an liveness sequence is constructed.
Step 3: and using the sensor record as a node, and constructing a sequence diagram based on the adjacent relation of the sensor sequence by using the adjacent record to correspond to the equal weight edge.
Step 4: and (3) constructing a graph signal based on the liveness based on the sensor record adjacent relation sequence graph.
Step 5: and calculating the frequency category of each node in the sequence chart.
Step 6: the sequence is partitioned based on the difference in frequency categories between the nodes connected by edges.
Further, in one embodiment of the present invention, converting the user location sub-sequence into a matrix representation based user location sub-sequence comprises: according to time sequence data of a plurality of training users, using a motion sensor as a node, and using the number of times that every two motion sensors are in an on state in adjacent records as an edge connecting the two nodes to construct a graph; determining an active area of a user to be tested according to the graph; mapping each user position of the user position sub-sequence into a corresponding active region to form an active region sequence; the sequence of active areas is converted into a binary adjacency matrix, resulting in a sub-sequence of user positions based on the matrix representation.
Specifically, the sequence representation adopts an adjacent matrix based on the active areas to represent potential relations among the sequences, the active areas of the users are obtained through clustering the sensors, the adjacent matrix is constructed according to the transfer times of the users among the active areas, and binarization is carried out so as to represent the relations among the sequences. As shown in fig. 6, the specific flow of the sequence representation is as follows:
step 1: the sequence representation sub-module receives a graph constructed by taking the motion sensor as a node and taking the number of times that each two motion sensors are in an on state in adjacent records of training set data as an edge connecting the two nodes, which is output by the data denoising module, and processes the output user position segmentation sequence based on the data segmentation sub-module.
Step 2: for motion sensor clustering, active regions are generated based on the generated clusters.
Step 3: and mapping each value in the segmented position sequence into an active area where the value is located, and constructing an active area sequence.
Step 4: a binary adjacency matrix is constructed based on the active area sequence.
Further, in an embodiment of the present invention, before acquiring the time series data of the user to be tested, the method further includes: training to obtain a mobile mode set according to time sequence data of a plurality of training users; specifically: determining a matrix representation of the position subsequence of the corresponding training user according to the time sequence data of the training users; determining a movement mode of the corresponding training user according to the matrix representation of the position subsequence of the corresponding training user; clustering is carried out according to the movement modes of a plurality of training users; and forming a moving pattern set according to the clustering result.
It can be understood that the embodiment of the invention further comprises a typical movement pattern mining function, receives the position subsequence information based on the matrix representation, and completes a typical template construction task and a dementia classification evaluation task based on the position subsequence information. When a template construction task is executed, clustering the position subsequence information sets of all users in the training set, and constructing a typical movement pattern set; when the dementia evaluation task is executed, mapping the position subsequence of the user to be tested into a movement pattern with the minimum distance from the matrix representation of the typical movement pattern set, and obtaining a movement pattern sequence of the user to be tested.
Specifically, the typical movement pattern mining module has different flows in performing the template building task and the dementia assessment task, including:
when the template construction task is executed, a typical movement pattern set is obtained by clustering binary adjacency matrixes of all user position segmentation sequences, and the specific flow is as follows as shown in fig. 7:
step 1: the typical movement pattern mining module receives a corresponding set of matrices representing sub-sequences of segment positions of all users of the training set as processed by the learning module.
Step 2: step 3-step 6 is performed for each combination of the splitting threshold and the typical cluster threshold within the preset splitting threshold search range and the typical cluster threshold search range.
Step 3: and generating an initial clustering result according to the splitting threshold value.
Step 4: and (5) adjusting the clustering result.
Step 5: the typical clusters and atypical clusters are divided. Counting the number of elements contained in all clusters, marking the clusters with the number of elements smaller than the threshold value of the typical cluster as atypical clusters, and marking the rest clusters as typical clusters.
Step 6: the extracted set of typical movement patterns is evaluated and a clustering result score is calculated.
Step 7: and selecting the optimal movement pattern set as a typical movement pattern set according to the clustering result score.
Step 8: a movement pattern sequence is calculated for each user in the training set.
Step 9: a typical movement pattern set and a movement pattern sequence of each user are output.
When performing dementia assessment tasks, mapping a matrix representation of a sequence of position segments of a user under test to a set of typical movement patterns, as shown in fig. 8, the specific flow is as follows:
step 1: the representative movement pattern mining module receives a corresponding matrix representation from the sub-sequence of segment locations of the user under test representing the processing output of the learning module, and a representative movement pattern set obtained by the representative movement pattern mining module upon receipt of the template construction task.
Step 2: and mapping the movement pattern to generate a movement pattern sequence of the user to be detected.
Step 3: and outputting a movement pattern sequence of the user to be tested.
In step S104, an activity pattern diagram of the user to be tested is determined according to the user position sub-sequence expressed based on the matrix, and a behavior pattern of the user to be tested is analyzed based on the activity pattern diagram.
In one embodiment of the present invention, determining an activity pattern diagram of a user to be tested according to a user position subsequence based on matrix representation includes: determining a target movement mode of a user to be detected from a movement mode set according to a user position subsequence based on matrix representation; according to the target movement mode, combining time sequence data of a user to be detected to construct an active mode primitive; and determining an activity pattern diagram of the user to be tested according to the activity pattern primitive.
It can be appreciated that when performing the template construction task, an activity pattern primitive is constructed by the typical movement pattern and the auxiliary sensor information, and an activity pattern diagram is constructed for each user in the training set; when the dementia evaluation task is executed, an activity pattern diagram of the user to be tested is generated based on the activity pattern primitive.
Further, in one embodiment of the invention, the categories of sensors include: auxiliary class, according to the goal moving pattern, combine the time series data of the user to be measured to construct the activity pattern primitive, including: and constructing an active mode primitive according to the target movement mode and the states of the auxiliary class sensors at each target time point in the time sequence data.
Specifically, the activity pattern construction module is used to construct an activity pattern graph of the user. By executing the template construction task, a corresponding activity pattern diagram can be generated for each user in the training set, as shown in fig. 9, and the specific steps are as follows:
step 1: the active pattern construction module receives a set of typical movement pattern sequences and a movement pattern sequence for each user.
Step 2: an active mode primitive is composed based on the typical movement pattern and auxiliary sensor state combination. All possible combinations of all typical movement patterns and all auxiliary sensor states constitute a full set of active pattern primitives. All users share the same active mode corpus. The user set is divided into a user training set and a user testing set.
Step 3: an activity pattern sequence is constructed for each user in the training set. The following operations are performed for each user in the training set: and sequentially acquiring the movement modes in the movement mode sequence of the user, mapping to corresponding active mode elements by combining the auxiliary sensor states, and arranging according to the original sequence to obtain the active mode sequence of the user.
Step 4: an activity pattern graph is constructed for each user in the training set. The following operations are performed for each user in the training set: and taking the active mode primitives as nodes, counting the transfer times between every two active mode primitives of a user, and taking the normalized transfer times as the weight of the edges between the corresponding active mode primitives.
Step 5: and outputting the activity pattern diagram and the activity pattern primitive of each user for the subsequent modules to use.
Further, when the dementia evaluating task is executed, an activity pattern diagram of the user to be tested is generated, and the activity pattern diagram of the user to be tested is output, as shown in fig. 10, the specific steps are as follows:
step 1: the active pattern construction module receives a set of typical movement patterns and a sequence of movement patterns for the user under test.
Step 2: and constructing a user activity mode sequence to be tested based on the user movement mode and the auxiliary sensor state.
Step 3: and constructing an activity pattern diagram of the user to be tested based on the relation among the primitives.
Step 4: and outputting an activity pattern diagram of the user to be tested to the illness state evaluation module.
Further, in one embodiment of the present invention, after determining the activity pattern diagram of the user to be tested according to the user location subsequence based on the matrix representation, the method further includes: determining a judging parameter of the active mode diagram; when used to analyze the pattern of user dementia activity, determining a first parameter of a pre-trained typical dementia template and determining a second parameter of a pre-trained typical healthy template; determining a first similarity between the activity pattern diagram and a typical dementia activity template and a second similarity between the activity pattern diagram and a typical health activity template according to the judging parameters, the first parameters and the second parameters; and analyzing the behavior mode of the user to be tested according to the first similarity and the second similarity.
It can be understood that the embodiment of the invention also has a dementia evaluation function, and when a template construction task is executed, a typical dementia movable template and a typical health movable template are generated, and an edge filtering threshold value and a judgment parameter of an activity pattern diagram are obtained; and when the dementia evaluation task is executed, the classification and dementia evaluation result of the user to be tested are given out by evaluating the similarity between the activity pattern diagram of the user to be tested and the typical activity template.
In one embodiment of the present invention, before obtaining the time series data of the user to be tested, the method further includes: according to time series data of a plurality of training users, a typical dementia movable template and a typical health movable template are obtained by combining the training of a movement pattern set, and specifically: determining an activity pattern diagram of each training user according to the time sequence data of each training user; and obtaining a typical dementia movable template and a typical health movable template by using the activity pattern diagram of each training user according to the dementia condition labels of each training user.
Specifically, the embodiment of the invention provides an evaluation method which classifies users as healthy or diseased and gives quantitative evaluation results. The disease evaluation module completes the disease evaluation method. Specifically, a typical healthy and a typical diseased active template are first generated, and then a user activity pattern diagram is compared with the typical healthy and diseased active templates, and a quantitative evaluation result for the user is given based on similarity with the two templates.
When performing the template building task, the dementia assessment function generates a typical healthy and a typical diseased active template, determines optimal parameters including, but not limited to: the characteristic weight of the edge, the optimal edge weight filtering threshold and the optimal judgment threshold. When the template construction task is executed, as shown in fig. 11, the specific steps are as follows:
step 1: an activity pattern diagram for each user in the training set is obtained.
Step 2: and constructing a healthy movable template and a diseased movable template.
Step 3: feature weights for each edge in the active pattern graph are calculated by minimizing the similarity of healthy and diseased active templates.
Step 4: and determining an edge filtering threshold searching range according to the healthy movable templates and the diseased movable templates, and executing the steps 5-7.
Step 5: and processing the user activity pattern diagram, the healthy activity template and the diseased activity template according to the side weight filtering threshold value to generate a temporary user activity pattern diagram, a temporary healthy activity template and a temporary diseased activity template, wherein the original user activity pattern diagram, the healthy activity template and the diseased activity template are not modified by the side filtering operation.
Step 6: based on the feature weight of each edge, the temporary activity pattern graph, the temporary healthy activity template and the temporary diseased activity template, the disease probability of each user in the training set is calculated.
Step 7: and calculating decision scores for all decision thresholds in the search range. For the decision threshold in the decision threshold search range, the following operations are performed: if the illness probability calculated in the step 5-6 of the user is larger than the judgment threshold, judging the user as an illness user, otherwise, calculating a judgment score according to the judgment evaluation index. The judgment evaluation index is a weighted average of the classification accuracy of the healthy users and the classification accuracy of the sick users, wherein the weight is a preset value and is related to the duty ratio of the healthy users and the sick users in the training set.
Step 8: and selecting an optimal side weight filtering threshold and an optimal judgment threshold according to the judgment score, and storing the optimal side weight filtering threshold and the optimal judgment threshold for a subsequent module to use. And selecting an edge weight filtering threshold value and a decision threshold value which enable the decision evaluation index to be maximum as an optimal edge weight filtering threshold value and an optimal decision threshold value.
Step 9: a typical healthy active template and a typical diseased active template are generated.
Step 10: and outputting a typical healthy movable template, a typical diseased movable template, an optimal side weight filtering threshold and an optimal judgment threshold.
Further, when the dementia evaluation task is executed, the dementia evaluation functional module obtains the dementia evaluation result of the user to be tested by calculating the similarity between the activity pattern diagram of the user to be tested and the typical activity template. When the dementia assessment task is performed, as shown in fig. 12, the specific steps are as follows:
Step 1: and acquiring an activity pattern diagram, an optimal side weight filtering threshold, an optimal judgment threshold, a typical healthy activity template and a typical diseased activity template of the user to be tested.
Step 2: and calculating the illness probability of the user to be detected, classifying the user, and classifying the user as the dementia user when the illness probability is larger than the optimal judgment threshold, otherwise, classifying the user as the healthy user.
Step 3: and outputting the dementia evaluation result of the user to be tested.
In summary, the embodiment of the invention carries out user behavior pattern learning and evaluation based on environmental sensor time series data, and provides a user behavior pattern learning method based on sensor time series data of graph signal processing and graph theory, and a dementia classification and evaluation method.
Specifically, the general flow of the present invention is shown in fig. 13 and 14, wherein (1) as shown in fig. 13, the flow of a typical template building task includes: inputting sensor time series data of a plurality of users formed by specific user executing specified activities, and dividing the data into a training set and a testing set; denoising training set sequence data of the sensor time sequence through a data denoising module to obtain a user position sequence; generating, by the representation learning module, a matrix-based representation of the sequence of segment positions; generating a typical movement pattern set through a typical movement pattern mining function module; an activity pattern diagram is built for each user in the training set through an activity pattern building functional module; and generating a typical dementia movable template and a typical health movable template through the dementia classification and evaluation functional module to obtain an edge filtering threshold value and a judgment parameter of the movable pattern diagram. (2) As shown in fig. 14, the flow of dementia classification and assessment tasks includes: inputting user sequence data to be detected, a typical health movable template, a typical dementia movable template and an optimal parameter set; denoising the user sensor sequence to be tested based on the data denoising functional module; generating a segmented position sequence representation of the user to be tested based on the matrix based on the representation learning module; mapping the matrix representation of the segmented position sequence of the user to be tested to a typical movement mode through a typical movement mode mining functional module; generating an activity pattern diagram of a user to be tested based on an activity pattern construction functional module; and calculating the similarity between the activity pattern diagram of the user to be tested and the typical template through the dementia evaluation module to obtain the classification and dementia evaluation result of the user to be tested.
In addition, the embodiment of the invention also comprises an initialization process, mainly comprising an initialization setting process of preset activities, an initialization process of sensor configuration, a training set generation process and an initial parameter setting process of a system. Wherein,
initialization setup procedure for preset activities initialization setup is performed for relevant parameters and procedures of daily activity procedure having several steps, including but not limited to: the daily activities need to have certain displacement behaviors or in-situ intermittent actions, and actions involved in the steps of the daily activities need to involve interaction with a plurality of home devices provided with sensors.
The initialization process of the sensor configuration refers to optimal deployment of a plurality of auxiliary sensors and motion sensors on equipment and articles needing interaction in an activity step, and the initialization process of the sensors is completed. The detection range of the motion detection sensor should cover a movement path for performing a preset activity, i.e., the user should be detected by the motion detection sensor when performing the activity.
The training set generation and initial parameter setting process refers to collecting sensor sequence data generated when a certain number of dementia patients and healthy users perform preset activities in an environment based on the preset activities and sensor configuration and the initialization process thereof.
The initial parameter setting process is a parameter initialization process required by the pointer-to-user activity pattern representation learning and dementia classification evaluation system, wherein the initialization parameters include, but are not limited to, a filter number search range in a denoising module, active state signal values and silent state signal values, a cluster splitting threshold search range, health illness weights in a decision evaluation index, a typical cluster threshold search range and a decision threshold search range.
According to the behavior pattern analysis method based on the Internet of things, based on the home environment where the environment sensor is deployed, the user performs activities in the environment according to the predefined steps, triggers the state change of the environment sensor, acquires the corresponding sensor state sequence, forms the sensor time sequence data of the user, denoises the sensor data based on the data, realizes automatic segmentation of the sequence, performs user behavior pattern recognition, reduces the cost of manually processing the data, improves the availability of sensor data analysis, performs dementia assessment from the aspects of qualitative and quantitative analysis, and can improve the interpretability of the data analysis.
Next, a behavior pattern analysis system based on the internet of things according to an embodiment of the present invention is described with reference to the accompanying drawings.
Fig. 15 is a block diagram of an internet of things-based behavior pattern analysis system according to an embodiment of the present invention.
As shown in fig. 15, the behavior pattern analysis system 10 based on the internet of things includes: the system comprises an acquisition module 100, a data denoising function module 200, a representation learning function module 300 and an activity pattern construction function module 400.
The acquiring module 100 is configured to acquire time-series data of a user to be tested, where the time-series data includes: the user to be tested executes at least one target time point of the preset activities according to the appointed steps, and sensor events corresponding to the target time point describe the state change condition of each sensor at the target time point; the data denoising function module 200 is used for determining a position sequence of a user to be detected according to the time sequence data; the representation learning function module 300 is configured to determine a sub-sequence of user positions based on a matrix representation based on the time series data in combination with the position sequence; the activity pattern constructing function module 400 is configured to determine an activity pattern diagram of the user to be tested according to the user position subsequence based on the matrix representation, and analyze a behavior pattern of the user to be tested based on the activity pattern diagram. The system 10 of the embodiment of the invention realizes automatic segmentation of the sequence based on the sensor data sequence by adopting graph wavelet transformation, reduces the cost of manually processing the data, improves the usability and the interpretability of data analysis based on the sensor time sequence by adopting fine granularity user activity pattern recognition, and reduces the invasion of user privacy and discomfort caused by monitoring based on a wearable sensor.
It will be appreciated that the system 10 of the present embodiment, as shown in fig. 16, includes a data denoising function module, a presentation learning function module, a typical movement pattern mining function module, an activity pattern construction function module, and a dementia condition assessment function module. Based on the functional module, the template construction task is mainly completed, and based on the constructed template, the dementia classification and evaluation task is completed. The typical mobile mode mining function module, the active mode construction module, and the dementia condition assessment module differ in flow when performing different tasks.
Further, in one embodiment of the present invention, the system 10 of the embodiment of the present invention further comprises: dementia assessment functional module. The dementia evaluation functional module is used for determining the judging parameters of the activity pattern graph after determining the activity pattern graph of the user to be tested according to the user position subsequence expressed based on the matrix, determining the first parameters of the pre-trained typical dementia activity template and determining the second parameters of the pre-trained typical health activity template when the dementia evaluation functional module is used for analyzing the dementia activity pattern of the user, and respectively determining the first similarity of the activity pattern graph and the typical dementia activity template and the second similarity of the activity pattern graph and the typical health activity template according to the judging parameters, the first parameters and the second parameters, and analyzing the behavior pattern of the user to be tested according to the first similarity and the second similarity.
Further, in one embodiment of the present invention, the representation learning function module 300 includes: the data segmentation sub-module and the sequence representation sub-module. The data segmentation sub-module is used for carrying out segmentation processing on the position sequence according to the time sequence data to obtain a user position sub-sequence; the sequence representation sub-module is for converting the user position sub-sequence into a matrix representation based user position sub-sequence.
Further, in one embodiment of the present invention, the active mode construction function module 400 is further configured to determine a target movement mode of the user to be tested from the movement mode set according to the user position sub-sequence based on the matrix representation, construct an active mode primitive according to the target movement mode in combination with time series data of the user to be tested, and determine an active mode graph of the user to be tested according to the active mode primitive.
Further, in one embodiment of the invention, the categories of sensors include: the auxiliary class, the activity pattern constructing function module 400 is further configured to construct an activity pattern primitive according to the target movement pattern in combination with the states of the auxiliary class sensor at each target time point in the time series data.
Further, in an embodiment of the present invention, the data denoising function module 200 is further configured to denoise the time-series data, so as to determine a location sequence of the user to be measured.
Further, in one embodiment of the invention, the categories of sensors include: the data denoising function module 200 is further configured to construct a graph by using, according to time series data of a plurality of training users, a sensor of the motion class as a node, the number of times that each two motion sensors are in an on state in adjacent records as a side connecting the two nodes, and performing graph wavelet transform on the graph signal by using a sensor state corresponding to each target time point in time series data of a user to be detected as a graph signal, to obtain a transform result corresponding to each target time point, determining, according to the transform result, an identifier of the target sensor corresponding to each target time point and serving as a user position identifier, wherein, on the target time point, the actual position distance between the target sensor corresponding to the target time point and the user to be detected is the smallest, and the category of the target sensor is the motion class, and form a position sequence according to the user position identifiers corresponding to the plurality of target time points.
Further, in one embodiment of the present invention, the system 10 of the embodiment of the present invention further comprises: a typical movement pattern digs functional modules. The system comprises a typical mobile mode mining function module, a mobile mode set generation module and a user detection module, wherein the typical mobile mode mining function module is used for obtaining a mobile mode set according to time sequence data of a plurality of training users before obtaining the time sequence data of the users to be detected; the dementia evaluation functional module is further used for obtaining a typical dementia movable template and a typical health movable template by combining the movement pattern set training according to the time series data of a plurality of training users.
Further, in one embodiment of the present invention, the typical mobile pattern mining function module is further configured to determine a matrix representation of a position sub-sequence of the respective training user according to the time-series data of each training user, determine a mobile pattern of the respective training user according to the matrix representation of the position sub-sequence of the respective training user, perform a clustering process according to the mobile patterns of the plurality of training users in combination with the matrix representation of the respective position sub-sequence, and form a mobile pattern set according to a result of the clustering process.
Further, in an embodiment of the present invention, the dementia evaluation function module is further configured to obtain an activity pattern diagram of a corresponding training user determined according to time-series data of each training user, and obtain a typical dementia activity template and a typical health activity template according to the dementia condition label of each training user by using the activity pattern diagram of each training user.
Further, in an embodiment of the present invention, the data segmentation submodule is further configured to construct an activity sequence for a state of a motion sensor of time series data, take all the sensors as nodes, take adjacent records as equal weight edges correspondingly, construct a sequence chart based on adjacent relations of the sensor sequence, construct a chart signal based on activity according to the activity sequence and the sequence chart of adjacent relations, calculate a frequency class of each node in the sequence chart based on the activity, calculate a frequency class difference between nodes connected by edges, and perform segmentation processing on the position sequence according to the chart signal based on activity.
Further, in an embodiment of the present invention, the sequence representation sub-module is further configured to construct a graph by using, as a side connecting two nodes, the number of times each two motion sensors are in an on state in an adjacent record according to time-series data of a plurality of training users, and the graph is used to determine an active area of a user to be tested, map each user position of the user position sub-sequence into a corresponding active area, form an active area sequence, and convert the active area sequence into a binary adjacency matrix, thereby obtaining a user position sub-sequence based on matrix representation.
It should be noted that the foregoing explanation of the embodiment of the behavior pattern analysis method based on the internet of things is also applicable to the behavior pattern analysis system based on the internet of things of the embodiment, which is not repeated herein.
According to the behavior pattern analysis system based on the Internet of things, based on the home environment where the environment sensor is deployed, the user performs activities in the environment according to the predefined steps, triggers the state change of the environment sensor, acquires the corresponding sensor state sequence, forms the sensor time sequence data of the user, denoises the sensor data based on the data, realizes automatic segmentation of the sequence, performs user behavior pattern recognition, reduces the cost of manually processing the data, improves the availability of sensor data analysis, performs dementia assessment from the aspects of qualitative and quantitative analysis, and improves the interpretability of the data analysis.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.
In the present invention, unless expressly stated or limited otherwise, a first feature "up" or "down" a second feature may be the first and second features in direct contact, or the first and second features in indirect contact via an intervening medium. Moreover, a first feature being "above," "over" and "on" a second feature may be a first feature being directly above or obliquely above the second feature, or simply indicating that the first feature is level higher than the second feature. The first feature being "under", "below" and "beneath" the second feature may be the first feature being directly under or obliquely below the second feature, or simply indicating that the first feature is less level than the second feature.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.

Claims (22)

1. The behavior pattern analysis method based on the Internet of things is characterized by comprising the following steps of:
obtaining time sequence data of a user to be tested, wherein the time sequence data comprises the following steps: the user to be tested executes at least one target time point of preset activities according to the appointed steps, and sensor events corresponding to the target time point describe the state change condition of each sensor at the target time point;
determining the position sequence of the user to be detected according to the time sequence data, wherein the method comprises the following steps: determining the identification of a target sensor corresponding to each target time point and taking the identification as a user position identification, wherein the category of the target sensor is a motion category, and forming the position sequence according to the user position identifications corresponding to a plurality of target time points;
combining the position sequences according to the time sequence data, and determining a user position subsequence based on matrix representation;
Determining a target movement mode of the user to be detected from a movement mode set according to the user position subsequence based on matrix representation;
according to the target movement mode, combining time sequence data of the user to be detected to construct an active mode primitive;
and determining an activity pattern diagram of the user to be tested according to the activity pattern primitive, and analyzing a behavior pattern of the user to be tested based on the activity pattern diagram.
2. The method for analyzing the behavior pattern based on the internet of things according to claim 1, wherein after determining the activity pattern diagram of the user to be tested according to the user position subsequence based on the matrix representation, the method further comprises:
determining a decision parameter of the activity pattern diagram;
when used to analyze the pattern of user dementia activity, determining a first parameter of a pre-trained typical dementia template and determining a second parameter of a pre-trained typical healthy template;
determining a first similarity between the activity pattern diagram and the typical dementia activity template and a second similarity between the activity pattern diagram and the typical health activity template according to the judging parameter, the first parameter and the second parameter;
And analyzing the behavior mode of the user to be tested according to the first similarity and the second similarity.
3. The behavioral pattern analysis method based on the internet of things according to claim 1, wherein said determining a user location subsequence based on a matrix representation based on said time-series data in combination with said location sequence comprises:
carrying out segmentation processing on the position sequence according to the time sequence data to obtain a user position subsequence;
the user position sub-sequence is converted into the matrix representation based user position sub-sequence.
4. The behavior pattern analysis method based on the internet of things according to claim 1, wherein the categories of the sensors include: auxiliary class, the said according to the said goal moving pattern, combine the time series data of the said subscriber to be measured to construct the activity pattern primitive, including:
and constructing the active mode primitive according to the target movement mode and the states of auxiliary sensors at each target time point in the time sequence data.
5. The behavioral pattern analysis method based on the internet of things according to claim 1, wherein the determining the position sequence of the user to be detected according to the time-series data comprises:
And denoising the time sequence data so as to determine the position sequence of the user to be detected.
6. The behavior pattern analysis method based on the internet of things according to claim 5, wherein the categories of the sensors include: and a motion class, wherein the denoising processing is performed on the time sequence data, so as to determine a position sequence of the user to be detected, and the motion class comprises the following steps:
according to time sequence data of a plurality of training users, using a motion sensor as a node, and using the number of times that every two motion sensors are in an on state in adjacent records as an edge connecting the two nodes to construct a graph;
taking the sensor states corresponding to each target time point in the time sequence data of the user to be detected as graph signals, and performing graph wavelet transformation on the graph signals to obtain transformation results corresponding to each target time point;
and determining the identifiers of the target sensors corresponding to the target time points according to the transformation result and taking the identifiers as user position identifiers, wherein the actual position distance between the target sensors corresponding to the target time points and the user to be detected is minimum at the target time points, and the user position identifiers corresponding to the target time points are used for forming the position sequence.
7. The behavioral pattern analysis method based on the internet of things according to claim 1, further comprising, before the obtaining of the time-series data of the user to be tested:
training to obtain the mobile mode set according to time sequence data of a plurality of training users;
and according to time series data of a plurality of training users, combining the movement pattern set to train to obtain a typical dementia movable template and a typical health movable template.
8. The method for analyzing a behavior pattern based on internet of things according to claim 7, wherein training the set of movement patterns according to time-series data of a plurality of training users comprises:
determining a matrix representation of a position subsequence of the corresponding training user according to the time sequence data of each training user;
determining a movement mode of the corresponding training user according to the matrix representation of the position subsequence of the corresponding training user;
clustering is carried out according to the movement modes of a plurality of training users;
and forming the moving pattern set according to the clustering result.
9. The behavioral pattern analysis method based on the internet of things of claim 7 wherein training in combination with the movement pattern set based on time-series data of a plurality of training users to obtain the typical dementia activity template and the typical health activity template comprises:
Determining an activity pattern diagram of the corresponding training user according to the time sequence data of each training user;
and obtaining the typical dementia movable templates and the typical health movable templates by using the activity pattern graphs of the training users according to the dementia condition labels of the training users.
10. The behavioral pattern analysis method based on the internet of things of claim 3 wherein said processing said sequence of locations in segments based on said time series data comprises:
constructing an activity sequence according to the state of the motion sensor of the time sequence data;
taking all sensor records as nodes, and correspondingly taking adjacent records as equal weight edges to construct a sequence diagram based on the adjacent relation of the sensor sequences;
according to the liveness sequence, constructing a liveness-based graph signal by combining the sequence graphs of the adjacent relations;
according to the activity-based graph signals, calculating the frequency category of each node in the sequence graph of the adjacent relation;
and carrying out segmentation processing on the position sequence based on the frequency class difference among the nodes connected by the edges.
11. The method for behavioral pattern analysis based on the internet of things of claim 10 wherein said converting said subsequence of user locations to said subsequence of user locations based on a matrix representation comprises:
According to time sequence data of a plurality of training users, using a motion sensor as a node, and using the number of times that every two motion sensors are in an on state in adjacent records as an edge connecting the two nodes to construct a graph;
determining an active area of the user to be tested according to the graph;
mapping each user position of the user position sub-sequence into a corresponding active region to form an active region sequence;
and converting the active region sequence into a binary adjacency matrix, thereby obtaining the user position subsequence based on matrix representation.
12. An internet of things-based behavior pattern analysis system, the system comprising:
the acquisition module is used for acquiring time sequence data of the user to be detected, wherein the time sequence data comprises the following components: the user to be tested executes at least one target time point of preset activities according to the appointed steps, and sensor events corresponding to the target time point describe the state change condition of each sensor at the target time point;
the data denoising function module is used for determining the position sequence of the user to be detected according to the time sequence data, wherein the determining the position sequence of the user to be detected according to the time sequence data comprises the following steps: determining the identification of a target sensor corresponding to each target time point and taking the identification as a user position identification, wherein the category of the target sensor is a motion category, and forming the position sequence according to the user position identifications corresponding to a plurality of target time points;
A representation learning function module for determining a user location sub-sequence based on a matrix representation in accordance with the time-series data in combination with the location sequence;
and the active mode construction function module is used for determining a target moving mode of the user to be tested from a moving mode set according to the user position subsequence based on matrix representation, constructing an active mode primitive according to the target moving mode and combining time sequence data of the user to be tested, determining an active mode diagram of the user to be tested according to the active mode primitive, and analyzing a behavior mode of the user to be tested based on the active mode diagram.
13. The behavior pattern analysis system based on the internet of things of claim 12, further comprising:
the dementia evaluation function module is used for determining the judging parameters of the activity pattern graph after determining the activity pattern graph of the user to be tested according to the user position subsequence based on the matrix representation, determining the first parameters of a pre-trained typical dementia activity template and determining the second parameters of a pre-trained typical health activity template when the dementia activity pattern is analyzed, and respectively determining the first similarity of the activity pattern graph and the typical dementia activity template and the second similarity of the activity pattern graph and the typical health activity template according to the judging parameters, the first parameters and the second parameters, and analyzing the behavior pattern of the user to be tested according to the first similarity and the second similarity.
14. The behavior pattern analysis system based on the internet of things of claim 12, wherein the representation learning function module comprises:
the data segmentation sub-module is used for carrying out segmentation processing on the position sequence according to the time sequence data to obtain a user position sub-sequence;
a sequence representation sub-module for converting the user position sub-sequence into the matrix representation based user position sub-sequence.
15. The internet of things-based behavior pattern analysis system of claim 12, wherein the categories of sensors include: and the activity mode construction function module is further used for constructing the activity mode primitive according to the target movement mode and the states of auxiliary class sensors at each target time point in the time sequence data.
16. The behavior pattern analysis system based on the internet of things according to claim 12, wherein the data denoising function module is further configured to denoise the time-series data, so as to determine the location sequence of the user to be detected.
17. The internet of things-based behavior pattern analysis system of claim 16, wherein the categories of sensors include: the data denoising function module is further used for constructing a graph by taking the sensor of the motion class as a node and the number of times that each two motion sensors are in an on state in adjacent records as the edge connecting the two nodes according to time sequence data of a plurality of training users, taking the sensor state corresponding to each target time point in the time sequence data of the user to be detected as a graph signal, carrying out graph wavelet transformation on the graph signal to obtain transformation results corresponding to each target time point, determining the identification of the target sensor corresponding to each target time point according to the transformation results and taking the identification as a user position identification, wherein on the target time point, the actual position distance between the target sensor corresponding to the target time point and the user to be detected is minimum, and the user position identifications corresponding to the target time points are used for forming the position sequence.
18. The behavior pattern analysis system based on the internet of things of claim 13, further comprising:
the typical mobile mode mining function module is used for training to obtain the mobile mode set according to the time sequence data of a plurality of training users before acquiring the time sequence data of the users to be tested;
the dementia evaluation functional module is further used for obtaining a typical dementia movable template and a typical health movable template by combining the movement pattern set training according to time series data of a plurality of training users.
19. The internet of things-based behavior pattern analysis system of claim 18, wherein the representative movement pattern mining function module is further configured to determine a matrix representation of a position sub-sequence of a corresponding training user based on time-series data of each training user, determine movement patterns of the corresponding training user based on the matrix representation of the position sub-sequence of the corresponding training user, perform a clustering process on movement patterns of a plurality of the training users, and form the movement pattern set based on a result of the clustering process.
20. The behavioral pattern analysis system of claim 18 further comprising means for obtaining an activity pattern of a respective training user determined from time-series data of each training user, and means for obtaining the representative dementia activity template and the representative health activity template from the activity pattern of each training user based on the dementia condition label of each training user.
21. The behavior pattern analysis system based on the internet of things according to claim 14, wherein the data segmentation submodule is further used for constructing an activity sequence for states of motion sensors of the time series data, taking all sensor records as nodes, correspondingly taking adjacent records as equal weight edges, constructing a sequence diagram based on adjacent relation of the sensor sequences, constructing a graph signal based on activity according to the activity sequence by combining the sequence diagram based on the adjacent relation, calculating frequency types of all nodes in the sequence diagram based on the activity, and carrying out segmentation processing on the position sequence based on frequency type differences among nodes connected by edges.
22. The behavior pattern analysis system based on internet of things according to claim 21, wherein the sequence representation sub-module is further configured to construct a graph with the motion sensor as a node and the number of times each two motion sensors are in an on state in adjacent records as an edge connecting the two nodes according to time series data of a plurality of training users, determine an active area of the user to be detected according to the graph, map each user position of the user position sub-sequence into a corresponding active area to form an active area sequence, and convert the active area sequence into a binary adjacency matrix, thereby obtaining the user position sub-sequence based on matrix representation.
CN202010380711.1A 2020-05-08 2020-05-08 Behavior pattern analysis method and system based on Internet of things Active CN111695426B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010380711.1A CN111695426B (en) 2020-05-08 2020-05-08 Behavior pattern analysis method and system based on Internet of things

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010380711.1A CN111695426B (en) 2020-05-08 2020-05-08 Behavior pattern analysis method and system based on Internet of things

Publications (2)

Publication Number Publication Date
CN111695426A CN111695426A (en) 2020-09-22
CN111695426B true CN111695426B (en) 2024-01-05

Family

ID=72477345

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010380711.1A Active CN111695426B (en) 2020-05-08 2020-05-08 Behavior pattern analysis method and system based on Internet of things

Country Status (1)

Country Link
CN (1) CN111695426B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103218442A (en) * 2013-04-22 2013-07-24 中山大学 Method and system for life mode analysis based on mobile device sensor data
WO2016163904A1 (en) * 2015-04-08 2016-10-13 Siemens Aktiengesellschaft Determining sensor positions in a supply network
CN108182382A (en) * 2017-12-06 2018-06-19 东南大学 Based on the similar Activity recognition method and system of figure
CN109620241A (en) * 2018-11-16 2019-04-16 青岛真时科技有限公司 A kind of wearable device and the movement monitoring method based on it

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103218442A (en) * 2013-04-22 2013-07-24 中山大学 Method and system for life mode analysis based on mobile device sensor data
WO2016163904A1 (en) * 2015-04-08 2016-10-13 Siemens Aktiengesellschaft Determining sensor positions in a supply network
CN108182382A (en) * 2017-12-06 2018-06-19 东南大学 Based on the similar Activity recognition method and system of figure
CN109620241A (en) * 2018-11-16 2019-04-16 青岛真时科技有限公司 A kind of wearable device and the movement monitoring method based on it

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Sparsity-Assisted Signal Denoising and Pattern Recognition in Time-Series Data;G.V. Prateek 等;http://arxiv.org/abs/1906.11330v1;全文 *
传感器网络时间序列数据的事件分类研究;叶李;;重庆邮电大学学报(自然科学版)(第03期);全文 *

Also Published As

Publication number Publication date
CN111695426A (en) 2020-09-22

Similar Documents

Publication Publication Date Title
Raykov et al. Predicting room occupancy with a single passive infrared (PIR) sensor through behavior extraction
US11902857B2 (en) Handling concept drift in Wi-Fi-based localization
Deep et al. A survey on anomalous behavior detection for elderly care using dense-sensing networks
US10607147B2 (en) Estimating a number of occupants in a region
JP5143212B2 (en) Method and apparatus for modeling behavior using probability distribution functions
Atallah et al. The use of pervasive sensing for behaviour profiling—a survey
Ge et al. Automatically detecting the small group structure of a crowd
Aminikhanghahi et al. Using change point detection to automate daily activity segmentation
EP1864246B1 (en) Spatio-temporal self organising map
Chidurala et al. Occupancy estimation using thermal imaging sensors and machine learning algorithms
Fan et al. Human fall detection using slow feature analysis
Zhang et al. ISEE Smart Home (ISH): Smart video analysis for home security
CN110136832B (en) Cognitive ability assessment system and method based on daily behaviors of old people
CN116602663B (en) Intelligent monitoring method and system based on millimeter wave radar
De Paola et al. User detection through multi-sensor fusion in an AmI scenario
Ye SLearn: Shared learning human activity labels across multiple datasets
Hu et al. Building occupancy detection and localization using cctv camera and deep learning
Yin et al. Unsupervised daily routine and activity discovery in smart homes
CN111695426B (en) Behavior pattern analysis method and system based on Internet of things
KR101340287B1 (en) Intrusion detection system using mining based pattern analysis in smart home
Mohmed et al. Convolutional neural network classifier with fuzzy feature representation for human activity modelling
KR20230117945A (en) System and method for analyzing customer object
CN113616209A (en) Schizophrenia patient discrimination method based on space-time attention mechanism
Tasmin et al. Comparative study of classifiers on human activity recognition by different feature engineering techniques
Fasching et al. Automated coding of activity videos from an OCD study

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant