CN112541621B - Movement prediction method, intelligent terminal and storage medium - Google Patents

Movement prediction method, intelligent terminal and storage medium Download PDF

Info

Publication number
CN112541621B
CN112541621B CN202011392382.9A CN202011392382A CN112541621B CN 112541621 B CN112541621 B CN 112541621B CN 202011392382 A CN202011392382 A CN 202011392382A CN 112541621 B CN112541621 B CN 112541621B
Authority
CN
China
Prior art keywords
data
movement
region
newly added
historical
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
CN202011392382.9A
Other languages
Chinese (zh)
Other versions
CN112541621A (en
Inventor
史文中
沈枭麒
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.)
Shenzhen Research Institute HKPU
Original Assignee
Shenzhen Research Institute HKPU
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 Shenzhen Research Institute HKPU filed Critical Shenzhen Research Institute HKPU
Priority to CN202011392382.9A priority Critical patent/CN112541621B/en
Publication of CN112541621A publication Critical patent/CN112541621A/en
Application granted granted Critical
Publication of CN112541621B publication Critical patent/CN112541621B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G06F18/24155—Bayesian classification

Abstract

The invention discloses a movement prediction method, an intelligent terminal and a storage medium, wherein the method comprises the following steps: acquiring newly added mobile data and historical mobile data, and clustering the newly added mobile data and the historical mobile data to generate a corresponding newly added interest area; when the similarity value between the newly added interest area and a preset historical interest area accords with a preset similarity rule, updating the historical moving data according to the newly added moving data to obtain updated moving data; training a preset initial movement prediction model according to the updated movement data to generate a target movement prediction model; and when the specified time to be predicted is detected, inputting the specified time into the target movement prediction model, generating a target area corresponding to the specified time and outputting the target area. When the method is used for carrying out the mobile prediction, the interest area is updated in real time by combining new data and old data so as to improve the accuracy of a subsequent prediction result.

Description

Movement prediction method, intelligent terminal and storage medium
Technical Field
The invention relates to the technical field of geographic information, in particular to a movement prediction method, an intelligent terminal and a storage medium.
Background
Research on movement of an individual is one of important problems in smart cities, social sciences, and the like, and has wide applications, such as providing weather forecast in real time according to movement of an individual. There are four broad categories of current personal movement-based prediction methods, including those based on Markov chains, those based on sequence models, those based on classification models, and those based on other models such as neural networks. The method based on the classification model has the advantages of comprehensively utilizing space-time characteristics and being suitable for sparse and dense data. However, the current prediction method based on the classification model is relatively fixed, and the consideration of the individual behavior change is lacked, for example, the following two points:
(1) change of active area: with the advance of time, an individual may change the range of the activity area due to moving, transferring and the like, including exploring a new activity area and leaving the old activity area, and the existing method cannot update the activity area in real time and detect the change of the activity area;
(2) change of movement law in the area: due to the reasons of time (such as seasons), traffic (road maintenance), personal preference and the like, the visiting time, visiting sequence and the like of an individual to an area may change, and the existing method cannot detect the change of the movement law.
Disclosure of Invention
The invention mainly aims to provide a movement prediction method, an intelligent terminal and a storage medium, and aims to solve the problem of inaccurate movement prediction caused by personal movement mode change in the prior art so as to improve the reliability and accuracy of the movement prediction.
In order to achieve the above object, the present invention provides a movement prediction method, including the steps of:
acquiring newly added mobile data and historical mobile data, and clustering the newly added mobile data and the historical mobile data to generate a corresponding newly added interest area;
when the similarity value between the newly added interest area and a preset historical interest area accords with a preset similarity rule, updating the historical moving data according to the newly added moving data to obtain updated moving data, wherein the historical interest area is an interest area obtained by clustering according to the historical moving data;
training a preset initial movement prediction model according to the updated movement data to generate a target movement prediction model;
and when the specified time to be predicted is detected, inputting the specified time into the target movement prediction model, generating a target area corresponding to the specified time and outputting the target area.
Optionally, the movement prediction method, wherein the training a preset initial movement prediction model according to the updated movement data to generate a target movement prediction model specifically includes:
carrying out de-duplication processing on the updated mobile data to obtain effective mobile data;
clustering the effective mobile data to generate an updated interest area corresponding to the updated mobile data;
taking the updating interest area corresponding to each effective mobile data as a corresponding current access area, and extracting the spatial feature and the time feature corresponding to each effective mobile data according to the updating interest area corresponding to each effective mobile data and the movement relation between the effective mobile data to obtain a movement event corresponding to each effective mobile data;
and training a preset initial prediction model according to the movement event to generate a target movement prediction model.
Optionally, the movement prediction method, wherein the updated movement data includes a collection time and a collection coordinate; the performing deduplication processing on the updated mobile data to obtain effective mobile data specifically includes:
sequencing the updated mobile data according to the sequence of the acquisition time to generate a mobile data sequence;
sequentially calculating the difference value of the acquisition time between the Nth updated moving data and the Mth updated moving data in the moving data sequence to obtain moving time, wherein N and M are positive integers less than or equal to the number of the updated moving data, and N is less than M;
when the moving time is less than or equal to a preset moving time threshold value, calculating a corresponding moving distance according to the acquisition coordinate between the Nth updated moving data and the Mth updated moving data;
and when the moving distance is smaller than or equal to a preset moving distance threshold value, taking the Nth updated moving data as effective moving data.
Optionally, the movement prediction method, wherein the spatial feature comprises a previously visited region; taking the update interest area corresponding to each effective mobile data as a corresponding current access area, and extracting the spatial feature and the temporal feature corresponding to each effective mobile data according to the update interest area corresponding to each effective mobile data and the movement relationship between the effective mobile data to obtain the movement event corresponding to each effective mobile data, which specifically includes:
sequencing each effective mobile data according to the sequence of the acquisition time of the effective mobile data to generate an effective mobile data sequence;
taking an updated interest region corresponding to the ith valid mobile data in the valid mobile data sequence as a previous access region corresponding to the (i + 1) th valid mobile data in the valid mobile data sequence, taking an updated interest region corresponding to the (i + 1) th valid mobile data in the valid mobile data sequence as a corresponding current access region, and generating a spatial feature corresponding to the ith valid mobile data and an updated interest region corresponding to the ith valid mobile data according to the previous access region and the current access region
And generating a time characteristic corresponding to each effective moving data according to a preset time division rule and the collection time of each effective moving data, wherein i is a positive integer smaller than the number of the effective moving data.
Optionally, in the movement prediction method, the time division rule includes a week division rule and a day division rule.
Optionally, the movement prediction method, wherein the initial prediction model comprises a feature weight object comprising the spatial feature and the temporal feature; the training a preset initial prediction model according to the movement event to generate a target movement prediction model specifically comprises:
for each feature weight object, calculating a feature training probability value of the feature weight object corresponding to each updated interest region according to the mobile event;
and taking the updated interest region corresponding to the training probability value with the maximum value as a feature training prediction region, and adjusting the weight value corresponding to the feature weight object according to the current access region in the mobile event, the feature training prediction region and a preset weight adjustment formula aiming at each mobile event to generate a target mobile prediction model.
Optionally, the movement prediction method, where the initial prediction model includes a plurality of candidate prediction models, and the training of the preset initial prediction model according to the movement event to generate the target movement prediction model specifically includes:
for each mobile event, calculating a model training probability value of the mobile event corresponding to each updating interest region based on the candidate prediction model, the time feature in the mobile event and a previous access region;
taking an updated interest region corresponding to the model training probability value with the largest value as a model training prediction region corresponding to the mobile event, and adjusting the weight value corresponding to each candidate model according to the current access region in the mobile event, the training prediction region and a preset weight adjustment formula to obtain a target weight value corresponding to each candidate model;
and taking the candidate model corresponding to the target weight value with the maximum value as the movement prediction model.
Optionally, the movement prediction method, wherein the weight adjustment formula
Figure BDA0002813184750000051
Wherein the object is a weight object comprising the candidate prediction model and the feature weight object, wobjectA weight value corresponding to each of the characteristic weight objects, c is a coefficient, and p is rmaxCorresponding training probability values including feature training probability values and model training probability values, rmaxFor training prediction regions, including feature prediction regions and model prediction regions, rtureIs the current access area in each of the movement events.
Optionally, the motion prediction method, wherein the candidate models comprise a naive bayes model with an adaptive window and a naive bayes model with an attenuation factor; based on the naive Bayes model with the attenuation factors, predicting the test probability value of the occurrence of the test event as
Figure BDA0002813184750000061
Figure BDA0002813184750000062
Wherein the test event is the spatial feature
Figure BDA0002813184750000063
The current access area is riAnd the time characteristic is
Figure BDA0002813184750000064
N is a positive integer of the number of spatial features, m is a positive integer of the number of temporal features;
rstfor said updating of the region of interest, riFor the currently visited area, x, in the test eventjFor the spatial feature, tkIs the time characteristic;
Figure BDA0002813184750000065
is characterized by the space
Figure BDA0002813184750000066
And the current access area is riIs determined by the sum of the counts corresponding to the movement events,
Figure BDA0002813184750000067
for the time characteristic to be tmAnd the current access area is riIs determined by the sum of the counts corresponding to the movement events,
Figure BDA0002813184750000068
for the current access area to be riIs determined by the sum of the counts corresponding to the movement events,
Figure BDA0002813184750000069
the sum of the counts corresponding to all movement events;
when the test event corresponds to the testWhen t is moment, the counting formula corresponding to each moving event is
Figure BDA00028131847500000610
io is the movement event and λ is a preset attenuation factor.
Optionally, the movement prediction method, wherein, when a similarity value between the new interest region and a historical interest region corresponding to the historical movement data meets a preset similarity rule, updating the historical movement data according to the new movement data to obtain updated movement data specifically includes:
when the number of the newly added mobile data is larger than or equal to a preset newly added data threshold value, clustering the newly added mobile data and the historical mobile data to generate a plurality of newly added interest areas;
according to each newly added interest area and each historical interest area, respectively carrying out area labeling on the newly added mobile data and the historical mobile data to obtain a newly added area data set corresponding to each newly added interest area and a historical area data set corresponding to each historical interest area;
calculating the region similarity value between each newly added region data set and each historical region data set aiming at each newly added region data set;
marking a newly added region data set and a historical interest region set corresponding to the region similarity value which is greater than or equal to a preset region similarity threshold value as a same region label;
calculating the overall similarity value between the newly added region data set and the historical interest region set according to the region label corresponding to each newly added region data set and each historical region data set;
and when the overall similarity value is smaller than a preset overall similarity value, updating the historical movement data according to the newly added movement data to obtain updated movement data.
Optionally, the movement prediction method, wherein before clustering the newly added movement data and the historical movement data to generate a plurality of newly added interest regions, when the number of the newly added movement data is greater than or equal to a preset newly added data threshold, the method further includes:
judging whether a historical interest region corresponding to the newly added mobile data exists or not according to the acquired coordinates of the newly added mobile data;
and if so, adjusting the weight value of the initial movement prediction model according to the newly added movement data.
In addition, to achieve the above object, the present invention further provides an intelligent terminal, wherein the intelligent terminal includes: a memory, a processor and a movement prediction program stored on the memory and executable on the processor, the movement prediction program when executed by the processor implementing the steps of the movement prediction method as described above.
In addition, to achieve the above object, the present invention further provides a storage medium, wherein the storage medium stores a movement prediction program, and the movement prediction program implements the steps of the movement prediction method as described above when executed by a processor.
Compared with the existing movement prediction method, the method for predicting the movement of the mobile terminal, provided by the invention, considers the change of the movement mode and comprises the following steps:
(1) in order to solve the problem of change of an activity area, after an interest area is extracted from historical data, new data is acquired each time, the interest area is updated in real time, and a similarity measurement method is provided for detecting whether the interest area changes;
(2) in order to overcome the problem of change of a movement rule in an activity area, a prediction model for detecting mode change and weakening influence of outdated data is provided, the model is based on a new classification model, and is combined with an existing stream data classification method, so that the stability of the prediction model is enhanced, and the characteristics and the weight of the model are updated in real time through a brand new method.
Through the two points, the embodiment of the invention can obtain a better prediction result of the movement of the prediction object.
Drawings
FIG. 1 is a flow chart of a preferred embodiment of a motion prediction method according to the present invention;
FIG. 2 is a schematic diagram of a model training process provided by the motion prediction method according to the present invention;
FIG. 3 is a diagram illustrating a process for updating historical movement data according to a preferred embodiment of the present invention;
FIG. 4 is a schematic diagram illustrating a change in motion pattern in a preferred embodiment of the motion prediction method of the present invention;
fig. 5 is a schematic operating environment diagram of an intelligent terminal according to a preferred embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, the movement prediction method according to the preferred embodiment of the present invention includes the following steps:
step S100, acquiring newly added mobile data and historical mobile data, and preprocessing the newly added mobile data and the historical mobile data to generate a corresponding newly added interest area.
Specifically, the mobile data refers to data related to the positioning of the predicted object, which is stored in advance and collected by the intelligent terminal, such as social media data, mobile phone signaling data, GPS positioning data, and the like. When the prediction object, that is, the user of the intelligent terminal, sends information through the intelligent terminal on the social network site, the current coordinates are obtained through Global Navigation Satellite System (GNSS), signaling, bluetooth, wifi, and the like, so as to display the current position of the prediction object. The mobile phone signaling has full coverage, and the base station can acquire the corresponding signal as long as the intelligent terminal is started, so that the mobile phone signaling is generated. Different base stations can be numbered in advance according to the mobile phone signaling data, and because the base station position coordinates of the base stations are fixed, after the position coordinates of each base station correspond to the base station numbers, when the mobile phone signaling is obtained, the corresponding base station position coordinates can be determined according to the corresponding base station numbers, and the base station position coordinates are used as the position coordinates corresponding to the intelligent terminal. Each piece of mobile data is composed of at least two parts, wherein one part is collection time, the other part is position coordinates, the collection time refers to the time when the mobile data is generated, and the position coordinates refer to the position of the intelligent terminal at the collection time. Besides the two parts, the number of the prediction object can be carried out on different intelligent terminals to identify different prediction objects, so that the historical mobile data of a plurality of people can be collected at the same time. And taking the historical movement data corresponding to the same prediction object number as a historical movement data set to perform subsequent movement prediction. For simplicity, the embodiment is described with a single prediction object.
In this embodiment, the movement data is divided into historical movement data and newly added movement data, the historical movement data is movement data used for determining the interest area at the previous time, and the newly added movement data is data related to the positioning of the intelligent terminal and collected after the historical interest area is determined. The interest region obtained by clustering the historical movement data is a historical interest region. And when the newly added moving data and the historical moving data are obtained, clustering the newly added moving data and the historical moving data according to the positioning information in the newly added moving data and the historical moving data, thereby generating a plurality of newly added interest areas corresponding to the newly added moving data and the historical moving data. It should be noted that the new movement data and the historical movement data of the clustering object are not clustered separately.
The clustering process can be realized by a clustering algorithm or a classification model of unsupervised learning. Taking a clustering algorithm as an example, the clustering algorithm classifies data into different categories according to the similarity between the data. The newly added movement data and the historical movement data are used as processing movement data, and the movement of a prediction object generally has a certain rule, for example, the working time is mainly concentrated in an office area. The processing movement data can thus be divided into groups by a clustering algorithm, describing the movement of the predicted object as a process from one group to another. Each processing movement data is represented as a point in space, so that the finally obtained processing movement data set can represent an area, and a prediction object appears in the area for multiple times, so that the prediction object is used as an interest area. Common Density-Based Clustering algorithms include a Noise-Based Density Clustering of Applications with Noise (DBSCAN), a Density Maximum Clustering algorithm (MDCA), and the like. However, DBSCAN requires two parameters to be determined, and it is difficult to select parameters adaptively, and multiple prediction objects cannot be adapted by using fixed parameters. Therefore, in the embodiment, Density-Based Noise application space Clustering (HDNSCAN) is preferably adopted, the HDNSCAN only needs one main parameter and is proved to be effective in an extraction interest area, and research reports also realize parameter self-adaptation according to the requirement of the number of clusters and the Clustering degree, so that the method is adopted to divide the processing mobile data into a plurality of processing mobile data sets. For each processing movement data set, generating a data area corresponding to the processing movement data set according to the acquired coordinates of the processing movement data in the processing movement data set, wherein the data area can include all the movement data in the corresponding processing movement data set, and the generated data area can adopt a convex hull algorithm and other modes. And finally, taking the data area as an updating interest area corresponding to each processing mobile data in the processing mobile data set.
Step S200, when the similarity value between the newly added interest area and a preset historical interest area accords with a preset similarity rule, updating the historical movement data according to the newly added movement data to obtain updated movement data, wherein the historical interest area is obtained by clustering according to the historical movement data.
Specifically, after historical movement data is obtained, clustering processing is performed on the historical movement data to obtain a corresponding historical interest area. The clustering processing method may adopt the above processing method capable of realizing the clustering effect, and is not described herein again. The similarity rule is a rule for measuring whether the newly added interest area is similar to the historical interest area. For example, a similarity threshold is preset for evaluating the similarity between the new interest region and the previous historical interest region. If the similarity value between the newly added interest area and the historical interest area corresponding to the historical movement data is smaller than a preset similarity value threshold value, the newly added interest area and the historical movement data are not similar, and therefore the newly added movement data generate large interference on the movement behavior of the prediction object. Therefore, the newly added movement data is added to the historical movement data to update the historical movement data to obtain updated movement data. The similarity value between the regions can be calculated according to the coincidence rate of the data corresponding to the two regions, or the area coincidence rate corresponding to the two regions can be adopted.
Further, to improve the accuracy of calculating the similarity between the newly added interest area and the historical movement data, step S200 includes:
step A10, when the number of the newly added mobile data is larger than or equal to a preset newly added data threshold, clustering the newly added mobile data and the historical mobile data to generate a plurality of newly added interest areas;
specifically, when the new movement data and the historical movement data are acquired, the historical interest area corresponding to the new movement data is calculated. The obtained new motion data includes the number, time and position of the prediction object, and the historical interest area corresponding to the new motion data is estimated based on the position of the new motion data, so that the new motion data may not belong to any historical interest area and is judged as noise.
And when the newly added mobile data reaches a preset newly added data threshold eta, clustering the newly added mobile data and the newly added mobile data to generate a plurality of newly added interest areas.
Further, when the number of the newly added mobile data is smaller than a preset newly added data threshold, that is, the newly added mobile data is not accumulated to a certain extent, the initial mobile prediction model may also be adjusted according to the newly added mobile data. The specific process is as follows:
judging whether an interest region corresponding to the newly added mobile data exists or not according to the acquisition coordinate of the newly added mobile data;
and if so, adjusting the weight value of the initial movement prediction model according to the newly added movement data.
The adjustment of the weight value of the initial prediction model is described in detail in the training of the initial prediction model, and is not described herein again.
Step A20, according to each new interest area and each historical interest area, performing area labeling on the new mobile data and the historical mobile data respectively to obtain a new area data set corresponding to each new interest area and a historical area data set corresponding to each historical interest area;
specifically, two groups of interest areas, namely a newly added interest area and a historical interest area, are added, and the historical interest area set is RSpA size of kpWherein the historical interest area is called ri(0≤i≤kp) (ii) a Newly added interest region set RScA size of kcWherein the newly added region of interest is called rj(0≤j≤kc). Because the interest areas corresponding to the newly added mobile data and the historical mobile data may be different, different area labels are performed on the newly added mobile data and the historical mobile data to obtain a newly added area data set and a historical area data set respectively.
For example, the history interest region corresponding to the newly added movement data a is region 1, and the corresponding newly added movement region is region 2. After the region labeling is performed, the data volume of all the newly added region data sets is equal to that of the historical region data sets. In the historical interest area or the new interest area, the new motion data serving as noise can be separately divided into a special interest area. For convenience of understanding, in the present embodiment, a new region of interest a, a new region of interest data B, a historical region of interest data C, and a new region data set a, a new region data set B, and a historical region data set C corresponding to the new region of interest a, the new region of interest data B, and the historical region data set C are described as an example.
Step A30, calculating the area similarity value between each new area data set and each history area data set;
specifically, the region similarity values between the newly added region data set a and the newly added region data set B and the history region data set C are calculated. The region similarity value can be calculated by adopting the coincidence rate of the areas and the similarity between the data sets. The present embodiment preferably adopts a calculation manner that is calculated according to the coincidence ratio of the movement data in the newly added region data set and the movement data in the history region data set.
The formula of calculation can be
Figure BDA0002813184750000141
Wherein
Figure BDA0002813184750000142
The intersection of the mobile data with the same geographic position determined according to the acquisition coordinates of each mobile data between the newly added area data set and the historical area data set,
Figure BDA0002813184750000143
the method comprises the steps that a union set of mobile data with the same geographic position is determined between a newly added area data set and a historical area data set according to the collection coordinates of all the mobile data; srIs the region similarity value.
Step A40, labeling a newly added region data set and a historical interest region set corresponding to the region similarity value which is greater than or equal to a preset region similarity threshold value as the same region label;
specifically, if the region similarity reaches a preset region similarity threshold, the two regions of interest are considered to be the same, for example, if the region similarity between the newly added region data set a and the history region data set C reaches the preset region similarity threshold, both the newly added region data set a and the history region data set C are labeled as region 1; and if the region similarity value between the newly added region data set B and the historical region data set C does not reach a preset region similarity threshold value, marking the newly added region data set B and the historical region data set C as different region labels, for example, marking the newly added region data set B as a region 2. This process may also be referred to as tag matching.
Step A50, calculating an overall similarity value between the newly added region data set and the historical interest region set according to the region label corresponding to each newly added region data set and each historical region data set;
specifically, after performing region label labeling on each newly added region data set and each history region data set, calculating an overall similarity value between the newly added region data set and the history interest region set. The number of the tags in the same area corresponding to the newly added area data set and the historical area data set can be calculated, and in this embodiment, the calculation formula can be
Figure BDA0002813184750000151
Wherein N isslIndicates that the number of the corresponding area labels of the newly added area data set and the historical area data set is the same, NlThe number of the newly added region data sets or the historical region data sets is referred to, if the similarity reaches a threshold value, the region of interest is considered to be unchanged, and if the similarity does not reach the threshold value, the region of interest is considered to be changed.
And step A60, when the overall similarity value is smaller than a preset overall similarity value, updating the historical movement data according to the newly added movement data.
Specifically, if the overall similarity value is greater than or equal to a preset overall similarity value, it is determined that the region of interest has not changed, and if the overall similarity value is smaller than the overall similarity value, it is determined that the region of interest has changed, so that the historical movement data is updated according to the newly added movement data, that is, the newly added movement data and the historical movement data are integrated to obtain updated movement data.
And step S300, training a preset initial movement prediction model according to the updated movement data to generate a target movement prediction model.
Specifically, according to the obtained updated movement data, a preset initial movement prediction model is trained to generate a target movement prediction model, wherein the initial movement prediction model may be a prediction model introduced in the background and based on personal movement prediction.
Further, in order to reduce the influence of redundant invalid data in updating the mobile data, the training process of the initial prediction model is also subjected to a deduplication processing, and the training process comprises the following steps:
step B10, carrying out de-duplication processing on the updated mobile data to obtain effective mobile data;
specifically, in order to reduce the influence of redundant invalid data in the updated moving data, a deduplication processing is also performed in the training process of the initial prediction model, and duplicated data in the updated moving data is deleted. For example, when the user is located in the updated interest area a at the time a and sends a message, one minute later, and the current time is the time b, the user is still located in the updated interest area a and sends a message, and on one hand, the data amount influences the calculation due to too large data amount, and on the other hand, the prediction result is not brought with a better effect. The redundant data can be deleted according to the change of the distance between the updated moving data at the similar acquisition time, so that the data calculation speed is improved.
Further, the process of the update moving data deduplication process adopted in this embodiment is as follows:
sequencing the updated mobile data according to the sequence of the acquisition time to generate a mobile data sequence;
sequentially calculating the difference value of the acquisition time between the Nth updated moving data and the Mth updated moving data in the moving data sequence to obtain moving time, wherein N and M are positive integers less than or equal to the number of the updated moving data, and N is less than M;
when the moving time is less than or equal to a preset moving time threshold value, calculating a corresponding moving distance according to the acquisition coordinate between the Nth updated moving data and the Mth updated moving data;
and when the moving distance is smaller than or equal to a preset moving distance threshold value, taking the Nth updated moving data as effective moving data.
Specifically, the acquisition time of all updated movement data is converted into seconds, and then the movement data is sequenced from small to large to generate a movement data sequence. Since an individual may send multiple pieces of information at the same location within a certain time, but the moving state thereof is not changed, deduplication is performed depending on time and location. Firstly, according to time, the difference value of the acquisition time between the Nth updated moving data and the Mth updated moving data in the moving data sequence is sequentially calculated to obtain moving time, and when the time interval of a plurality of pieces of updated moving data is smaller than a threshold value, whether the moving distance of the updated moving data is smaller than a preset moving distance threshold value is further checked. The moving distance between the nth updated moving data and the mth updated moving data can be calculated according to the acquired coordinates between the nth updated moving data and the mth updated moving data. Since the collected coordinates in this embodiment may have different forms, and the geodetic coordinate system is not a planar coordinate system and is similar to an ellipsoid, it is inconvenient to calculate the distance based on the geodetic coordinate system, and therefore, a standard planar coordinate system is preset for calculating the movement distance, and then the collected coordinates are projected to the planar coordinate system, and the distance between the collected coordinates and the planar coordinate system is calculated according to the projected coordinate values. If the moving distance is smaller than the moving distance threshold value, the plurality of pieces of updated moving data are considered to be repeated, and the earliest time among the plurality of pieces of data, namely the Nth updated moving data, is reserved, and the others are deleted as repeated data.
And step B20, clustering the effective movement data to generate an updated interest area corresponding to the updated movement data.
Specifically, this step is similar to the clustering process described above, and is not described herein again.
And step B30, according to the updated interest areas corresponding to the effective mobile data and the movement relation among the effective mobile data, taking the updated interest areas corresponding to the effective mobile data as corresponding current access areas, and extracting the spatial features and the temporal features corresponding to the effective mobile data to obtain the movement events corresponding to the effective mobile data.
In particular, the spatial feature is a feature indicating that valid movement data has changed in moving spatially, such as a distance between a previous access area, a forward access area, and a current access area; the time characteristic is a characteristic that valid movement data changes in time by moving, such as a time difference between a previous access area and a current access area, which week the moment of movement occurred, day or night in the day, and the like. Through the combination of the spatial features and the temporal features, a movement event corresponding to valid movement data can be obtained, and the movement event can be defined as predicting that an object accesses a certain updating interest area at a certain moment.
Further, the movement of the predicted object can be generally predicted according to previous behaviors, for example, if a certain predicted object passes through a bookstore, the certain predicted object may go to a library with a certain probability. Thus, in the present embodiment, the spatial variation of the movement event can be defined as<ra,rb>Wherein r isaFor predicting a history of interest region visited before the movement of the object, i.e. a previously visited region (previous visit), rbThe method is used for predicting the historical interest area visited after the object moves, namely the current visit area. Thus, step a30 includes:
sequencing each effective mobile data according to the sequence of the acquisition time of the effective mobile data to generate an effective mobile data sequence;
taking an updated interest region corresponding to the ith valid mobile data in the valid mobile data sequence as a previous access region corresponding to the (i + 1) th valid mobile data in the valid mobile data sequence, taking an updated interest region corresponding to the (i + 1) th valid mobile data in the valid mobile data sequence as a corresponding current access region, and generating a spatial feature corresponding to the ith valid mobile data and an updated interest region corresponding to the ith valid mobile data according to the previous access region and the current access region
And generating a time characteristic corresponding to each effective moving data according to a preset time division rule and the collection time of each effective moving data, wherein i is a positive integer smaller than the number of the effective moving data.
The time division rules comprise week division rules and day division rules. The day division rule is a time interval in a day, and refers to a time interval in a day when the prediction object accesses a certain update interest area. The embodiment adopts an existing time period dividing method, which specifically comprises the following steps: early morning (0: 00-5: 59), early morning (6:00-9:59), morning (10:00-13:59), afternoon (14:00-17:59), evening (18:00-20:59), late night (21:00-23: 59). The weekly partition rule is a partition rule for a day of the week, which refers to which day of the week the time of the object accessing a certain update interest area is, and this feature is used because the object exhibits different movement patterns on weekends and weekdays and is somewhat different even on weekdays. In addition, the user can set up a month division rule for obtaining the month in which the movement event occurs, a day and night division rule for obtaining the day or night in which the movement event occurs, and the like, so that the specific form of the time characteristic and the included types can be various.
And step B40, training a preset initial prediction model according to the movement event to generate a target movement prediction model.
Specifically, according to the movement event, a preset initial prediction model is trained, for example, time characteristics of the movement event are used as input values, the initial prediction model calculates a historical interest region corresponding to the time characteristics, then space characteristics of the movement time are used as real values, the predicted historical interest region is compared with the real values, parameter adjustment is performed on the initial prediction model according to the difference between the predicted historical interest region and the real values, and the target movement prediction model is generated until all movement events are trained.
Further, in this embodiment, the initial prediction model includes a feature weight object, and the feature weight object includes the spatial feature and the temporal feature; thus, the specific training process may be:
step C10, for each feature weight object, calculating a feature training probability value of the feature weight object corresponding to each updated interest region according to the movement event.
Specifically, taking the basic model of the initial prediction model as a naive bayes classifier as an example for description, the initial prediction model can be expressed as
Figure BDA0002813184750000201
Wherein r isstFor the current access area, pv is the previous access area, tm is a time characteristic obtained according to a day division rule, and dw is a time characteristic obtained according to a week division rule. w is apv,wtm,wdw,wrRespectively representing the weight values corresponding to the four characteristic weight objects. The larger the weight value, the more important the feature weight object is in the model. If an eigen weight object makes the final output result more inclined to the truth value r of the predicted resulttrueThen the feature weight object is important. Therefore, for each feature weight object, according to the movement event, a feature training probability value of the feature weight object corresponding to each updated interest region is calculated.
As can be seen from the above formula, there are four weight values w to be trainedpv,wtm,wdw,wrRespectively correspond to P (pv | r)st)，P(tm|rst)，P(dw|rst)，P(rst). With P (pv | r)st) For example, assuming that the movement event used for training is the predicted object transfer from region A to region B, then the value of pv is region A, rstIs the true value r corresponding to the moving event of the updated region of interest to be predictedtrueIs B, thus rstTaking different values, P (pv | r)st) There will be a corresponding probability value, which is used as a feature training probability value.
And step C20, taking the updated interest region corresponding to the feature training probability value with the maximum value as a feature training prediction region, and adjusting the weight value corresponding to the feature weight object according to the current access region in the mobile event, the feature training prediction region and a preset weight adjustment formula aiming at each mobile event to generate a target mobile prediction model.
Specifically, the probability value P (pv | r) is trained by selecting the featurest) Maximum rstAs a training prediction region rmax. Then according to rmaxAnd the current access area r of the moving timetrueAdjusting the weight wpv. The other three weight adjustments are the same.
For example, the input movement event is Monday, the previous access area is area A, the current access area is area B, and the pv is known to be area A, r according to the movement eventtrueFor region B, it is necessary to predict which update region of interest is the current region of interest in the input motion event. For example, if the weight value to be adjusted is wpvThen, when the current visited area takes different values, the probability that the visited area is the area a is calculated, for example, when the current visited area is the area B, the feature training probability value P (area a | area B) is 2/3, when the current visited area is the area C, the feature training probability value P (area a | area C) is 1/3, and the current visited area corresponding to the maximum value in the feature training probability values is selected as rmaxI.e. rmaxIs region B, so rtrue＝rmax(ii) a If the feature training probability value P (region A | region B) is 1/3 and the feature training probability value P (region A | region C) is 2/3, then region C is selected as rmaxAt this time rtrueIs not equal to rmax. If r ismaxAnd rtrueSimilarly, the feature weight object is considered to have a positive effect on the formula, and the weight of the feature weight object is increased, that is, the weight value corresponding to monday is increased; if not, the weight value is correspondingly reduced. The adjustment of the weight value can be realized according to a preset weight adjustment formula.
The present embodiment provides a weight adjustment formula:
Figure BDA0002813184750000221
wherein the object is a weight object comprisingThe candidate prediction model and the feature weight object, wobjectA weight value corresponding to each of the characteristic weight objects, c is a coefficient, and p is rmaxCorresponding training probability values including feature training probability values and model training probability values, rmaxFor training prediction regions, including feature prediction regions and model prediction regions, rtureIs the current access area in each of the movement events. When p is greater than 0.5, the result is more certain, and if it equals the true value, the weight is expanded more. And when p is less than 0.5, even if it is equal to the true value, the weight increase is small because it is not reliable enough. If the result is false and not equal to the true value, then whether p is greater than 0.5 or not is considered unreliable and the weight is directly reduced.
Further, the initial prediction model in this embodiment includes a plurality of candidate prediction models; thus, the specific training process may be:
for each mobile event, calculating a model training probability value of the mobile event corresponding to each updating interest region based on the candidate prediction model, the time feature in the mobile event and a previous access region;
taking an updated interest region corresponding to the model training probability value with the largest value as a model training prediction region corresponding to the mobile event, and adjusting the weight value corresponding to each candidate model according to the current access region in the mobile event, the training prediction region and a preset weight adjustment formula to obtain a target weight value corresponding to each candidate model;
and taking the candidate model corresponding to the target weight value with the maximum value as the movement prediction model.
Specifically, in order to obtain a better prediction effect, the initial prediction model may include a plurality of candidate prediction models, and a weight value may be set for each candidate model. And then training each candidate prediction model by adopting a mobile event, and adjusting the weight value corresponding to the candidate prediction model according to the accuracy of the prediction result to obtain a target weight value. When the model is used, the candidate model corresponding to the target weight value with the largest value is used as the movement prediction model. This process is similar to the above process of adjusting the weight corresponding to the feature weight object, and is not described herein again. The updating mode enables the target weight value of the candidate model with higher prediction precision in the near future to be larger. The upper bound of the target weight value is set to 1 and the lower bound is set to 0.1 instead of 0, so that the weight can reach the lower bound within a limited number of times and grow rapidly.
Further, in this embodiment, the candidate predictive models include a naive Bayes model with adaptive windows (A)
Figure BDA0002813184750000233
Bayes model with adaptive windowing, NB-ADWIN) and naive Bayes model with attenuation factor (A)
Figure BDA0002813184750000234
Bayes model with decay factors，NB-DF)。
Wherein, the basic model formula corresponding to NB-ADWIN is
Figure BDA0002813184750000231
Figure BDA0002813184750000232
In order to calculate each probability value in the formula, a plurality of windows are established for data required by each probability calculation, each window is used for detecting whether the corresponding data distribution is consistent, if not, the corresponding data before the acquisition time, namely the old data, is removed, the data with consistent distribution is reserved, and then the data is used for probability calculation. As the time changes, the movement pattern of the prediction object changes, for example, the interest area a is frequently visited on monday before, and then the interest area a is frequently visited on monday after, and the interest area a is most frequently visited before, but the interest area B is frequently visited due to moving and the like. Therefore, the prediction results guided by the newly added movement data and the historical movement data are separated, and the difference is reducedAnd errors, namely, data which are not uniform with the newly added moving data in the historical moving data are removed through the uniformity of the data, so that the interference of the historical moving data is reduced, and the accuracy of prediction is improved.
In addition, NB-ADWIN is a window-based technique that has certain disadvantages in that windows may not be able to adapt when the movement pattern changes relatively slowly. The slow change of the moving mode refers to that the regular change of the moving characteristic is slow, which may be that the moving characteristic such as moving distance, visiting sequence of the position, visiting time and the like changes gradually along with the change of the time, for example, moving from 1km to 1.1km to 1.2, 1.3, 1.4, 1.5km and the like every day, and the numerical value is slow to change, which includes the similar situation including but not limited to slow change of other characteristics, which may be called that the moving mode changes slowly. Furthermore, due to the uncertainty of the user movement, false detections may also result, so that a stable method that can attenuate the effects of outdated data is necessary. Based on this, the present embodiment proposes a naive Bayes model with attenuation factors (A)
Figure BDA0002813184750000244
Bayes models with decay factors, NB-DF). For each input test event, calculating a corresponding test probability value as:
the basic formula of the model is
Figure BDA0002813184750000241
Figure BDA0002813184750000242
Wherein the test event is the spatial feature
Figure BDA0002813184750000243
The current access area is riAnd the time characteristic is
Figure BDA0002813184750000251
N is the positive of the number of spatial featuresAn integer, m being a positive integer of the number of temporal features;
rstfor said updating of the region of interest, riFor the currently visited area, x, in the test eventjFor the spatial feature, tkIs the time characteristic;
Figure BDA0002813184750000252
is characterized by the space
Figure BDA0002813184750000253
And the current access area is riIs determined by the sum of the counts corresponding to the movement events,
Figure BDA0002813184750000254
for the time characteristic to be tmAnd the current access area is riIs determined by the sum of the counts corresponding to the movement events,
Figure BDA0002813184750000255
for the current access area to be riIs determined by the sum of the counts corresponding to the movement events,
Figure BDA0002813184750000256
is the sum of the counts for all movement events.
Typically each mobility event occurrence is counted as 1, but the NB-DF takes into account the effect on the prediction result of the length of time between the moment the test event occurs and the moment the respective mobility event occurs. In this embodiment, the sum of various types of movement events can be expressed as:
Figure BDA0002813184750000257
wherein io is the movement event, SioIs the set of all the mobile events io, t is the occurrence time corresponding to the test time, Wio,tIndicating that a count corresponding to the occurrence of a movement event io at time t,
Figure BDA0002813184750000258
λ is a preset attenuation factor, λ ∈ (0, 1). With the moment of occurrence of a movement event, i.e. tioThe length of time between the occurrence of the test event, i.e., (t-t)io) Since the influence of the movement event on the prediction result corresponding to the test event is smaller, the influence of the data with longer time should be reduced when the result is predicted by calculating the occurrence number of the event.
Further, when the time characteristics include a plurality of types of characteristics, such as the week characteristics divided according to the week division rule and the day characteristics of the day division rule, the test probability value calculated according to the test time may be represented as P (r)st＝ri|pv＝vpv,td＝vtd,dw＝vdw) Where pv is the previous visit area, td is the day feature, dw is the week feature, v is the previous visit areatdTo test the daily characteristics of an event, vdwTo test the weekly profile of the event. Therefore, the basic formula of the model can be adjusted correspondingly, for example, the basic formula of the NB-DF model is adjusted to
Figure BDA0002813184750000261
Wherein
Figure BDA0002813184750000262
Is at vtdAccess riThe number of movement events of (a) is,
Figure BDA0002813184750000263
at vdwAccess riThe number of movement events.
And step S400, when the specified time to be predicted is detected, inputting the specified time into the target movement prediction model, and generating and outputting a target area corresponding to the specified time.
In particular, the target movement prediction model is used to predict historical regions of interest to which the predicted object is most likely to move at a particular time. Therefore, when the appointed time to be predicted is detected, the appointed time is input into the target movement prediction model, the target movement prediction model calculates the probability value corresponding to each historical interest region according to the appointed time, and the historical interest region with the maximum probability value is used as the target region and is output.
Further, the main body performing the movement prediction method may be a plug-in installed on the software or may be packaged separately as a separate software. In addition, the movement prediction method can be combined with multiple functions, for example, weather forecast, a corresponding target area is predicted according to the working hours of a target user in advance aiming at a certain target user, and then the weather of the target area at the working hours is pushed to the target user accurately by combining the weather forecast, so that the target user can prepare clothes and rain gear in advance. Besides weather forecast, traffic real-time report, real-time early warning and the like can be combined with a movement prediction program and a movement prediction method, so that accurate pushing is realized.
To illustrate the prediction effect of the method for predicting movement of a prediction object according to the embodiment of the present invention, fig. 4 shows the position distribution of a user, the position coordinates of the user are divided into two groups by taking a certain time as a boundary, the movement data obtained before the time is classified into a first group, and the movement data obtained after the time is classified into a second group, wherein the position coordinates of the first group of movement data are marked by a circle, the position coordinates of the second group of movement data are marked by a square, the regions A, B, C and D framed by an oval frame and a square frame are four manually marked active regions, it can be seen from the figure that the difference between the first group of movement data and the second group of movement data is large, only the circle mark of the first group of movement data is present in the active region A, B, C, and only the square mark of the next group of movement data is present in the active region D, so that the interest preference of the user before and after the time is greatly changed, if the interest area is not updated, the interest area possibly existing in the range of the activity area D cannot be extracted, and only when an updating strategy exists, the new interest area can be extracted and is more suitable for the actual situation.
To further illustrate the advantages of the prediction object movement prediction method provided by the embodiment of the present invention compared with other mainstream methods, the following table shows the precision comparison between the NB-AWDF of the present method and the mainstream method Multi-feature Weighted Bayesian Model (MWBM) and the Sparse mobile Markov Chain Model (SMMC), where the precision is the ratio of the prediction correct times to the prediction total times, and η is the newly added data threshold. From the table, it can be seen that the precision of the method is up to 70.70% in the social networking site Instagram data, up to 59.95% in the social networking site Twitter data, 6.45% and 7.34% in the two data, 5.66% and 6.39% in average, 8.73% and 7.91% in the two data, and 7.91% and 7.54% in average, respectively, relative to the SMMC. Therefore, the prediction method for predicting the movement of the object provided by the embodiment of the invention has advantages compared with other mainstream methods.
Figure BDA0002813184750000281
Further, as shown in fig. 5, based on the above-mentioned movement prediction method, the present invention also provides an intelligent terminal, which includes a processor 10, a memory 20 and a display 30. Fig. 5 shows only some of the components of the smart terminal, but it is to be understood that not all of the shown components are required to be implemented, and that more or fewer components may be implemented instead.
The memory 20 may be an internal storage unit of the intelligent terminal in some embodiments, such as a hard disk or a memory of the intelligent terminal. The memory 20 may also be an external storage device of the Smart terminal in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are provided on the Smart terminal. Further, the memory 20 may also include both an internal storage unit and an external storage device of the smart terminal. The memory 20 is used for storing application software installed in the intelligent terminal and various data, such as program codes of the installed intelligent terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, the memory 20 stores a movement prediction program 40, and the movement prediction program 40 can be executed by the processor 10 to implement the movement prediction method of the present application.
The processor 10 may be, in some embodiments, a Central Processing Unit (CPU), a microprocessor or other data Processing chip, which is used for running program codes stored in the memory 20 or Processing data, such as executing the motion prediction method.
The display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch panel, or the like in some embodiments. The display 30 is used for displaying information at the intelligent terminal and for displaying a visual user interface. The components 10-30 of the intelligent terminal communicate with each other via a system bus.
In one embodiment, the following steps are implemented when the processor 10 executes the movement prediction program 40 in the memory 20:
acquiring newly added mobile data and historical mobile data, and clustering the newly added mobile data and the historical mobile data to generate a corresponding newly added interest area;
when the similarity value between the newly added interest area and the historical interest area corresponding to the historical movement data meets a preset similarity condition, updating the historical movement data according to the newly added movement data to obtain updated movement data;
training a preset initial movement prediction model according to the updated movement data to generate a target movement prediction model;
and when the specified time to be predicted is detected, inputting the specified time into the target movement prediction model, generating a target area corresponding to the specified time and outputting the target area.
The present invention also provides a storage medium, wherein the storage medium stores a movement prediction program, and the movement prediction program realizes the steps of the movement prediction method as described above when executed by a processor.
Of course, it will be understood by those skilled in the art that all or part of the processes of the methods of the above embodiments may be implemented by a computer program instructing relevant hardware (such as a processor, a controller, etc.), and the program may be stored in a computer readable storage medium, and when executed, the program may include the processes of the above method embodiments. The storage medium may be a memory, a magnetic disk, an optical disk, etc.
It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.

Claims (12)

1. A movement prediction method, characterized in that the movement prediction method comprises:
acquiring newly added mobile data and historical mobile data, and clustering the newly added mobile data and the historical mobile data to generate a corresponding newly added interest area;
when the similarity value between the newly added interest area and a preset historical interest area accords with a preset similarity rule, updating the historical moving data according to the newly added moving data to obtain updated moving data, wherein the historical interest area is an interest area obtained by clustering according to the historical moving data;
training a preset initial movement prediction model according to the updated movement data to generate a target movement prediction model; the training a preset initial movement prediction model according to the updated movement data to generate a target movement prediction model specifically includes:
carrying out de-duplication processing on the updated mobile data to obtain effective mobile data;
clustering the effective mobile data to generate an updated interest area corresponding to the updated mobile data;
taking the updating interest area corresponding to each effective mobile data as a corresponding current access area, and extracting the spatial feature and the time feature corresponding to each effective mobile data according to the updating interest area corresponding to each effective mobile data and the movement relation between the effective mobile data to obtain a movement event corresponding to each effective mobile data;
training a preset initial prediction model according to the movement event to generate a target movement prediction model;
and when the specified time to be predicted is detected, inputting the specified time into the target movement prediction model, generating a target area corresponding to the specified time and outputting the target area.
2. The movement prediction method according to claim 1, characterized in that the update movement data comprises a collection time and a collection coordinate; the performing deduplication processing on the updated mobile data to obtain effective mobile data specifically includes:
sequencing the updated mobile data according to the sequence of the acquisition time to generate a mobile data sequence;
sequentially calculating the difference value of the acquisition time between the Nth updated moving data and the Mth updated moving data in the moving data sequence to obtain moving time, wherein N and M are positive integers less than or equal to the number of the updated moving data, and N is less than M;
when the moving time is less than or equal to a preset moving time threshold value, calculating a corresponding moving distance according to the acquisition coordinate between the Nth updated moving data and the Mth updated moving data;
and when the moving distance is smaller than or equal to a preset moving distance threshold value, taking the Nth updated moving data as effective moving data.
3. The movement prediction method according to claim 1, characterized in that the spatial features comprise previously visited areas; taking the update interest area corresponding to each effective mobile data as a corresponding current access area, and extracting the spatial feature and the temporal feature corresponding to each effective mobile data according to the update interest area corresponding to each effective mobile data and the movement relationship between the effective mobile data to obtain the movement event corresponding to each effective mobile data, which specifically includes:
sequencing each effective mobile data according to the sequence of the acquisition time of the effective mobile data to generate an effective mobile data sequence;
taking an updated interest region corresponding to the ith valid mobile data in the valid mobile data sequence as a previous access region corresponding to the (i + 1) th valid mobile data in the valid mobile data sequence, taking an updated interest region corresponding to the (i + 1) th valid mobile data in the valid mobile data sequence as a corresponding current access region, and generating a spatial feature corresponding to the ith valid mobile data and an updated interest region corresponding to the ith valid mobile data according to the previous access region and the current access region
And generating a time characteristic corresponding to each effective moving data according to a preset time division rule and the collection time of each effective moving data, wherein i is a positive integer smaller than the number of the effective moving data.
4. The movement prediction method according to claim 3, characterized in that the time division rule includes a week division rule and a day division rule.
5. The movement prediction method according to claim 1, characterized in that the initial prediction model comprises an eigenweight object comprising the spatial feature and the temporal feature; the training a preset initial prediction model according to the movement event to generate a target movement prediction model specifically comprises:
for each feature weight object, calculating a feature training probability value of the feature weight object corresponding to each updated interest region according to the mobile event;
and taking the updated interest region corresponding to the feature training probability value with the maximum value as a feature training prediction region, and adjusting the weight value corresponding to the feature weight object according to the current access region in the mobile event, the feature training prediction region and a preset weight adjustment formula aiming at each mobile event to generate a target mobile prediction model.
6. The movement prediction method according to claim 5, wherein the initial prediction model includes a plurality of candidate prediction models, and the training of the preset initial prediction model according to the movement event to generate the target movement prediction model specifically includes:
for each mobile event, calculating a model training probability value of the mobile event corresponding to each updating interest region based on the candidate prediction model, the time feature in the mobile event and a previous access region;
taking an updated interest region corresponding to the model training probability value with the largest value as a model training prediction region corresponding to the mobile event, and adjusting the weight value corresponding to each candidate model according to the current access region in the mobile event, the training prediction region and a preset weight adjustment formula to obtain a target weight value corresponding to each candidate model;
and taking the candidate model corresponding to the target weight value with the maximum value as the movement prediction model.
7. The movement prediction method according to claim 6, wherein the weight adjustment formula
Figure FDA0003101716700000041
Wherein the object is a weight object comprising the candidate prediction model and the feature weight object, wobjectA weight value corresponding to each of the characteristic weight objects, c is a coefficient, and p is rmaxCorresponding training probability values including feature training probability values and model trainingProbability value rmaxFor training prediction regions, including feature prediction regions and model prediction regions, rtureIs the current access area in each of the movement events.
8. The method of motion prediction according to claim 6, characterized in that the candidate models comprise a naive Bayes model with adaptive windows and a naive Bayes model with attenuation factors; based on the naive Bayes model with the attenuation factors, predicting the test probability value of the occurrence of the test event as
Figure FDA0003101716700000042
Figure FDA0003101716700000051
Wherein the test event is the spatial feature
Figure FDA0003101716700000052
The current access area is riAnd the time characteristic is
Figure FDA0003101716700000053
N is a positive integer of the number of spatial features, m is a positive integer of the number of temporal features;
rstfor said updating of the region of interest, riFor the currently visited area, x, in the test eventjFor the spatial feature, tkIs the time characteristic;
Figure FDA0003101716700000054
for said spatial feature to be vxjAnd the current access area is riIs determined by the sum of the counts corresponding to the movement events,
Figure FDA0003101716700000055
for the time characteristic to be tmAnd the current access area is riIs determined by the sum of the counts corresponding to the movement events,
Figure FDA0003101716700000056
for the current access area to be riIs determined by the sum of the counts corresponding to the movement events,
Figure FDA0003101716700000057
the sum of the counts corresponding to all movement events;
when the test time corresponding to the test event is t, the counting formula corresponding to each mobile event is
Figure FDA0003101716700000058
io is the movement event and λ is a preset attenuation factor.
9. The movement prediction method according to claim 5, wherein when the similarity value between the new interest area and the historical interest area corresponding to the historical movement data meets a preset similarity rule, updating the historical movement data according to the new movement data to obtain updated movement data, specifically comprising:
when the number of the newly added mobile data is larger than or equal to a preset newly added data threshold value, clustering the newly added mobile data and the historical mobile data to generate a plurality of newly added interest areas;
according to each newly added interest area and each historical interest area, respectively carrying out area labeling on the newly added mobile data and the historical mobile data to obtain a newly added area data set corresponding to each newly added interest area and a historical area data set corresponding to each historical interest area;
calculating the region similarity value between each newly added region data set and each historical region data set aiming at each newly added region data set;
marking a newly added region data set and a historical interest region set corresponding to the region similarity value which is greater than or equal to a preset region similarity threshold value as a same region label;
calculating the overall similarity value between the newly added region data set and the historical interest region set according to the region label corresponding to each newly added region data set and each historical region data set;
and when the overall similarity value is smaller than a preset overall similarity value, updating the historical movement data according to the newly added movement data to obtain updated movement data.
10. The movement prediction method according to claim 9, wherein before clustering the newly added movement data and the historical movement data to generate a plurality of newly added interest regions when the number of the newly added movement data is greater than or equal to a preset newly added data threshold, the method further comprises:
judging whether a historical interest region corresponding to the newly added mobile data exists or not according to the acquired coordinates of the newly added mobile data;
and if so, adjusting the weight value of the initial movement prediction model according to the newly added movement data.
11. An intelligent terminal, characterized in that, intelligent terminal includes: memory, a processor and a movement prediction program stored on the memory and executable on the processor, the movement prediction program when executed by the processor implementing the steps of the movement prediction method according to any of claims 1-10.
12. A storage medium, characterized in that the storage medium stores a movement prediction program, which when executed by a processor implements the steps of the movement prediction method according to any one of claims 1 to 10.
CN202011392382.9A 2020-12-02 2020-12-02 Movement prediction method, intelligent terminal and storage medium Active CN112541621B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011392382.9A CN112541621B (en) 2020-12-02 2020-12-02 Movement prediction method, intelligent terminal and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011392382.9A CN112541621B (en) 2020-12-02 2020-12-02 Movement prediction method, intelligent terminal and storage medium

Publications (2)

Publication Number Publication Date
CN112541621A CN112541621A (en) 2021-03-23
CN112541621B true CN112541621B (en) 2021-08-31

Family

ID=75015450

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011392382.9A Active CN112541621B (en) 2020-12-02 2020-12-02 Movement prediction method, intelligent terminal and storage medium

Country Status (1)

Country Link
CN (1) CN112541621B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113450872B (en) * 2021-07-02 2022-12-02 南昌大学 Method for predicting phosphorylation site specific kinase
CN114328472B (en) * 2022-03-15 2022-05-27 北京数腾软件科技有限公司 AI-based data migration method and system

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104239556A (en) * 2014-09-25 2014-12-24 西安理工大学 Density clustering-based self-adaptive trajectory prediction method

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107977734B (en) * 2017-11-10 2021-08-24 河南城建学院 Prediction method based on mobile Markov model under space-time big data
GB2583718A (en) * 2019-05-01 2020-11-11 Samsung Electronics Co Ltd Method, apparatus and computer program for updating a cluster probability model
CN110149595B (en) * 2019-05-10 2021-01-08 北京工业大学 HMM-based heterogeneous network user behavior prediction method
CN111988744B (en) * 2020-08-31 2022-04-01 重庆邮电大学 Position prediction method based on user moving mode

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104239556A (en) * 2014-09-25 2014-12-24 西安理工大学 Density clustering-based self-adaptive trajectory prediction method

Also Published As

Publication number Publication date
CN112541621A (en) 2021-03-23

Similar Documents

Publication Publication Date Title
Chen et al. Understanding ridesplitting behavior of on-demand ride services: An ensemble learning approach
CN107111794B (en) Predicting and exploiting variability of travel time in a mapping service
EP3035314B1 (en) A traffic data fusion system and the related method for providing a traffic state for a network of roads
Rong et al. Du-parking: Spatio-temporal big data tells you realtime parking availability
CN112700072B (en) Traffic condition prediction method, electronic device, and storage medium
US10496881B2 (en) PU classifier for detection of travel mode associated with computing devices
Zhang et al. Modeling pedestrians’ near-accident events at signalized intersections using gated recurrent unit (GRU)
CN110414732B (en) Travel future trajectory prediction method and device, storage medium and electronic equipment
CN111080029B (en) Urban traffic road speed prediction method and system based on multi-path segment space-time correlation
CN110268454B (en) Determining a customized safe speed for a vehicle
CN107103753A (en) Traffic time prediction system, traffic time prediction method, and traffic model establishment method
US20160125307A1 (en) Air quality inference using multiple data sources
Fang et al. FTPG: A fine-grained traffic prediction method with graph attention network using big trace data
CN112541621B (en) Movement prediction method, intelligent terminal and storage medium
CN111582559B (en) Arrival time estimation method and device
CN108986453A (en) A kind of traffic movement prediction method based on contextual information, system and device
US10401181B2 (en) Detection of travel mode associated with computing devices
Rompis et al. Probe vehicle lane identification for queue length estimation at intersections
Li et al. Robust inferences of travel paths from GPS trajectories
Liang et al. NetTraj: A network-based vehicle trajectory prediction model with directional representation and spatiotemporal attention mechanisms
Chen et al. A multiscale-grid-based stacked bidirectional GRU neural network model for predicting traffic speeds of urban expressways
Koster et al. Recognition and recommendation of parking places
Gao et al. CTTE: Customized travel time estimation via mobile crowdsensing
CN114141385A (en) Early warning method and system for infectious diseases and readable storage medium
CN111812689A (en) User behavior analysis method and device based on GPS track, electronic equipment and medium

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