A clustering prediction method for target action areas based on temporal network model

Abstract

The present invention discloses a target action area clustering prediction method based on a temporal network model, comprising the following steps: preprocessing raw situation data; establishing a target behavior profile based on business rules and specific situation data content, and determining target behavior characteristics; mining the predicted target situation data using the DBSCAN algorithm, and calculating the behavior characteristics of all targets using a clustering method; establishing a target behavior prediction network model based on an LSTM structure based on the number of features in the feature set, and determining a training loss function for the network model; training the network model using the SGD algorithm, and exporting the model weights and network structure to a file for storage after training convergence; importing the trained model weights, applying them to real-time target data, and predicting the target's action or area. The present invention has high prediction accuracy and real-time performance, and can quickly identify and predict the target's action trends.

Description

Technical Field

[0001] The present invention belongs to the technical field of target situation data analysis, and in particular relates to a target action area clustering prediction method based on a temporal network model. Background Art

[0002] In target situational data analysis, predicting target behavior is key to improving situational awareness and combat effectiveness in adversarial environments. Due to the uncertainty of enemy target behavior patterns, enemy targets may adopt different tactics and strategies, and their behavior patterns will change with the dynamics of the adversarial environment and the dynamics of both sides. Therefore, accurately capturing and understanding enemy target behavior patterns is a challenging task. Furthermore, the diversity and complexity of adversarial data present challenges in predicting and analyzing target behavior. Battlefield data includes a variety of types, including behavioral data, historical records, and environmental factors. Effectively integrating and utilizing this data, extracting key features, and building predictive models are crucial for situational awareness and analysis.

[0003] Existing research on predicting target action intentions based on adversarial situation data primarily relies on two types of data: historical target behavior trajectory data and target characteristic data. Target action location prediction based on historical trajectory data typically utilizes target recognition models, target databases, and geographic environment databases to detect the target's specific location during its actions in real time. This allows for the development of a hidden Markov model that accurately predicts the target's action route and purpose. This model is then used to infer information such as the potential behavioral intentions of individual targets. In recent years, deep learning, with its advantages of efficient multidimensional data processing, flexible implementation, and strong versatility, has also begun to be applied to the prediction of individual target behavior destinations. For example, techniques such as BP neural networks have been used to improve the accuracy of predicting individual target behavior intentions.

[0004] Driven by artificial intelligence and machine learning technologies, target situation analysis is becoming increasingly important in numerous fields, including warfare, security, and urban planning. In complex scenarios, traditional analysis methods are no longer sufficient to accurately predict and respond to target movement areas. Therefore, leveraging advanced data mining techniques and intelligent algorithms to analyze and predict target movement areas has become increasingly urgent and necessary. Summary of the Invention

[0005] To achieve the above objectives, the target action area clustering prediction method based on the temporal network model disclosed in this application includes the following steps:

[0006] S1: Preprocessing of raw situation data, including data screening and data cleaning;

[0007] S2: Build a target behavior profile based on business rules and specific situation data content to determine the target behavior characteristics;

[0008] S3: Use the DBSCAN algorithm to mine the predicted target situation data and calculate the behavioral characteristics of all targets through clustering. The clustering results form a feature dataset about the target, which is used to train the subsequent neural network model.

[0009] S4: Based on the number of features in the feature set, a target behavior prediction network model based on the LSTM structure is established, and the training loss function of the network model is determined;

[0010] S5: Train the network model based on the SGD algorithm. After the training converges, export the model weights and network structure to a file for saving.

[0011] S6: Import the trained model weights, apply them to real-time target data, and predict the target's actions or areas.

[0012] Furthermore, the original situation data is described mathematically as:

[0013] ;

[0014] In the above formula, Indicates the target No. The spatiotemporal characteristic value of individual action, Indicates the target No. The time period when the spatiotemporal characteristics of similar behaviors occur, Indicates the target No. The spatiotemporal characteristics of class behaviors correspond to the key area numbers, Indicates the target No. The spatiotemporal characteristics of class behaviors correspond to the frequency of historical behaviors. Respectively represent the target No. The spatiotemporal characteristics of the action-like objects correspond to the longitude and latitude coordinates of key area points. Indicates the target No. The spatiotemporal characteristics of the action-like objects correspond to the action route numbers;

[0015] Target The longitude and latitude coordinates of the center or key points of the historical action key area .

[0016] Furthermore, the data screening is carried out by comparing target intelligence information, the geographical location of key areas and relevant business rules, screening out all target situation data in all historical periods, eliminating behavioral data related to non-targets of interest, retaining data records with target ID numbers, and eliminating data records without clear target IDs;

[0017] The data cleaning cleans up the repeated fields in the behavior records to ensure the uniqueness of each behavior record; and all types of information in each behavior record remain complete, otherwise it will be directly deleted as abnormal data; at the same time, behavior information that is not related to the analysis of the spatiotemporal characteristics of individual target behaviors is deleted.

[0018] Furthermore, step S3 further includes:

[0019] S31: Mark all historical behavior records of each target The initial state is True;

[0020] S32: Randomly extract a behavior record , replace its mark to False: Draw a circle with the center point and traverse the behavior data set corresponding to the target of the behavior record All objects marked as True in ;

[0021] S33: Determine the behavior record Attributes: If Centered on The historical behavior frequency within the circle with radius , then mark the behavior record as the cluster center and create the corresponding cluster , take the target recording time in the cluster The average value is the standard time of the cluster , and then randomly select the next behavior record , turn to S32 until the target All historical behavior records All records are traversed, that is, all records are marked as False, and the process goes to S34;

[0022] S34: If the target If the total number of marked center points is 0, it means that the target has no fixed behavior pattern and directly goes to S38. If the total number of marked center points is not 0, proceed to S35;

[0023] S35: For any two center points in a cluster, if the time interval between their records is less than EPS, the two points are connected. The connected center points and all the points in their neighborhood form a combined cluster. If a center point is not connected to any other center point, then the center point and all the points in its neighborhood form a cluster. The historical behavior records that do not form a cluster are recorded as noise points, indicating the random behavior of individual targets.

[0024] S36: Output individual goals All clusters generated:

[0025] ;

[0026] S37: All clusters Inner key area With coordinate set Corresponding matching, we can get the spatiotemporal characteristics of individual behavior, where is the track segment number generated based on the target action key point set:

[0027]

[0028] Then integrate the output target The spatiotemporal characteristic matrix of individual behavior;

[0029] ;

[0030] S38: Input the historical behavior record set of the next target ID and go to S31 until all target IDs are traversed and the algorithm ends.

[0031] Furthermore, the target behavior prediction network model based on the LSTM structure includes an encoding network, an LSTM network and a decoding network;

[0032] The encoding network structure consists of a 2-layer DNN neural network. The input of the model is the feature vector of the above target:

[0033] ;

[0034] Output the encoded intermediate vector , as the input of the subsequent LSTM network layer;

[0035] The LSTM network uses the target's time series behavior data to analyze the input; is the input information at time t, is the hidden vector of the network input at time t. The LSTM network recursively calls itself and passes the information at time t−1 to time t, taking the feature vector given by the encoding network as The LSTM network finally integrates a coding representation c of the current target state, which is used as the input of the subsequent decoding layer network.

[0036] The decoding network consists of a fully connected network, whose output is the input of the LSTM network, and outputs the probability distribution of the next key area or key point of the target ,The decoding network generates a specific label probability distribution based on the encoding representation c of the current state of the target, and selects the label with the largest probability as the prediction result.

[0037] Furthermore, the target behavior feature prediction model inputs the previous feature vectors into the network model in chronological order each time it makes a prediction, and the model finally gives the probability distribution of the next key area or key point of the target. , use the actual key points or key areas in the dataset during training One-Hot encoding representation of As labels, calculate the loss function:

[0038] ;

[0039] During training, the SGD optimizer is used to train on the target feature set to obtain the parameters of the model.

[0040] Compared with the existing target situation analysis method, the target action area clustering prediction method based on the time series network model proposed in the present invention has higher prediction accuracy and real-time performance. It can quickly identify and predict the target's action trend in massive data, thereby making more scientific and reasonable decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flowchart of an embodiment of the present invention;

[0042] Figure 2 is a flowchart of a process for analyzing target action intention according to an embodiment of the present invention;

[0043] Figure 3 This is a flow chart of the DBSCAN mining algorithm used in an embodiment of the present invention;

[0044] Figure 4 is a coded network data structure provided by an embodiment of the present invention;

[0045] Figure 5 It is the LSTM network data structure provided by the embodiment of the present invention;

[0046] Figure 6 This is a decoding network data structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention fall within the scope of protection of the present invention.

[0048] refer to Figure 1 and Figure 2 The target action area clustering prediction method based on the temporal network model disclosed in this application includes the following steps:

[0049] S1: Preprocessing of raw situation data, including data screening and data cleaning;

[0050] S2: Build a target behavior profile based on business rules and specific situation data content to determine the target behavior characteristics;

[0051] S3: Use the DBSCAN algorithm to mine the predicted target situation data and calculate the behavioral characteristics of all targets through clustering. The clustering results form a feature dataset about the target, which is used to train the subsequent neural network model.

[0052] S4: Based on the number of features in the feature set, a target behavior prediction network model based on the LSTM structure is established, and the training loss function of the network model is determined;

[0053] S5: Train the network model based on the SGD algorithm. After the training converges, export the model weights and network structure to a file for saving.

[0054] S6: Import the trained model weights, apply them to real-time target data, and predict the target's actions or areas.

[0055] In one embodiment, the target behavior intention prediction process based on the LSTM neural network is as follows:

[0056] Input data: target The current behavior state sequence to be predicted: ,Target The longitude and latitude coordinates of the center or key points of the historical action key area The meanings of the behavior status information fields are as follows:

[0057] (1) , indicating the target The action time number,

[0058] (2) , indicating the target The action area number,

[0059] (3) , indicating that the target The action track number;

[0060] Output: , the target's next possible action area number, the overall data processing flow can be found in Figure 2 .

[0061] Target situation spatiotemporal profiling is based on the labeling of land, sea, and air situational data related to the target situation. Labels are typically highly refined, manually defined feature systems. The set of labels corresponding to each target abstracts the target's complete information landscape, collectively forming a holistic description of the target. In other words, target profiling is a data analysis tool that accurately mines target characteristics and describes target behavior. Building a target profile generally involves three steps: target analysis, system construction, and profile creation. The core task is establishing a labeling system that reflects target attributes. Applying this concept to air situational analysis gives rise to the concept of target behavior spatiotemporal profiling. Target behavior spatiotemporal profiling refers to the labeling of target dynamic behavior information.

[0062] Target behavior information primarily relies on the Land, Sea, and Air Situation Information System (LSIS) and the geographic location of key areas and points. This system can acquire and store real-time dynamic behavior data for individual targets, including the longitude and latitude coordinates associated with each target ID, or action records at key points. This environmental geographic data, combined with deployment information from both sides, can be used to effectively quantify the characteristics of target action points.

[0063] This situational information cannot be directly used for data analysis and model training, and it is impossible to intuitively analyze the spatiotemporal patterns of target behavior. Therefore, it is necessary to label the land, sea, and air situational data and the geographic location information of key areas or key points. In other words, to construct a spatiotemporal profile of target behavior based on this land, sea, and air situational data. Accordingly, constructing a spatiotemporal profile of target behavior primarily involves three steps: preprocessing target behavior information, extracting spatiotemporal characteristic values ​​of target behavior, and visualizing the target behavior profile.

[0064] The spatiotemporal characteristics of target behavior in adversarial environment situation data refer to the statistical regularity exhibited by individual targets when acting within the same historical period and spatial interval. The spatiotemporal characteristics of target behavior often vary significantly. To facilitate quantitative research and intuitive expression, the spatiotemporal characteristics of target behavior based on real-time monitoring data from various reconnaissance equipment and geographic environmental information in key areas can be mathematically described as follows:

[0065] ;

[0066] In the above formula, Indicates the target No. The spatiotemporal characteristic value of individual action, Indicates the target No. The time period when the spatiotemporal characteristics of the behavior occur (generally accurate to the minute, specifically defined according to the business scenario), Indicates the target No. The spatiotemporal characteristics of class behaviors correspond to the key area numbers, Indicates the target No. The spatiotemporal characteristics of class behaviors correspond to the frequency of historical behaviors. Respectively represent the target No. The spatiotemporal characteristics of the action-like objects correspond to the longitude and latitude coordinates of key area points. Indicates the target No. The spatiotemporal characteristics of the action class correspond to the action route number.

[0067] The spatiotemporal behavior characteristics of a single target may also be different each time it performs a task, so there may be multiple types of individual spatiotemporal behavior characteristics in the historical action records of the target. The spatiotemporal characteristic matrix of individual actions It can be expressed as follows:

[0068] ;

[0069] The sources of target behavior information in confrontation scenarios include two categories: situation information acquired and integrated by the target reconnaissance and detection system and corresponding geographic location information.

[0070] Target situation data primarily includes complete information about individual targets, including track information, target type, and target movement direction. Because data anomalies and errors may occur during the storage of massive amounts of data, and the redundant data irrelevant to the study of individual target behavior patterns can also affect data analysis efficiency, the raw situation data set needs to be preprocessed. In one embodiment, the specific preprocessing rules are as follows:

[0071] (1) Data screening: By comparing target intelligence information, the geographical location of key areas, and relevant business rules, all target situation data within all historical periods are screened out. Behavioral data related to non-targets such as civilian targets need to be eliminated. Data records with target ID numbers are retained, and data records without clear target IDs are eliminated.

[0072] (2) Data cleaning: If all fields in two or more behavior records are repeated, the corresponding duplicate data needs to be cleaned to ensure the uniqueness of each behavior record; and all types of information in each behavior record must be complete, otherwise it will be directly eliminated as abnormal data; at the same time, behavior information that is not related to the analysis of the spatiotemporal characteristics of individual target behaviors, such as device numbers, also needs to be eliminated.

[0073] In summary, the valid data fields for obtaining target behavior information based on situation data are shown in Table 1. By matching and connecting these two types of data, the spatiotemporal characteristics of individual behavior targets can be fully represented.

[0074] Table 1 data valid fields:

[0075]

[0076] To accurately construct spatiotemporal profiles of individual behaviors, it is crucial to effectively extract reliable spatiotemporal characteristics of target individual behaviors. Because the number of spatiotemporal characteristics of target individual behaviors is unknown and uncertain, clustering methods that require a predetermined number of clusters are not suitable for extracting these characteristics. However, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), a density clustering method based on densely connected regions, does not require a predefined number of initial cores or clusters. Here, we use the DBSCAN algorithm to extract spatiotemporal characteristics of target individual behaviors, which serve as input to the subsequent spatiotemporal action prediction model.

[0077] The DBSCAN algorithm is a density-based, noisy clustering method. It divides a sufficiently dense spatial range into clusters. It can aggregate arbitrary clusters in a noisy data space and defines a cluster as the largest set of densely connected points. This algorithm assumes that the classification of research objects can be determined by their distribution density, and similar research objects are grouped together.

[0078] The algorithm employs two global parameters: 1) the maximum density distance (EPS), which is the radius of the neighborhood surrounding a target point; and 2) the minimum number of similar points (MinPts), which is the minimum number of points within the neighborhood surrounding a target point. Both parameters can be adjusted to suit the context and have no fixed values. Based on these two parameters, all points in the spatial database can be categorized into three types: 1) Center points, which are points whose neighborhood contains at least the same number of other points as the target point; 2) Edge points, which are points within the neighborhood of a center point but contain fewer than the target point; and 3) Noise points, which are points that are neither center points nor edge points.

[0079] This application applies the DBSCAN algorithm to extract the spatiotemporal characteristic values ​​of target individual behavior, mainly based on land, sea and air situation data, and takes the spatiotemporal behavior of individual targets as the research object. In this method, EPS is used to represent the ideal time interval between the target's actions at key points or key areas, so as to measure whether the target's actions between different key points (or key areas) have temporal regularity; MinPts represents the minimum number of activities of the target in a certain period of time or a certain key area, which is used to determine whether the target's behavior has spatial regularity. If the historical behavior frequency of the target in a certain key area within a certain period of time is , then the target's behavior in this area during this period is considered random. On the contrary, if , it is considered that the target moves according to a fixed rule in this spatial area during this period.

[0080] Input data: (1) ideal time interval EPS and minimum number of actions MinPts; (2) longitude and latitude coordinates of the center of the key area or key points in the situation data ; (3) Target set in situation data: ; (4) All historical behavior records of the target ,in, 、 Target No. The action time and key point (key area) number corresponding to each behavior record, .

[0081] Output: target Behavioral spatiotemporal characteristic matrix ;

[0082] .

[0083] refer to Figure 3 In one embodiment, the processing flow of the improved DBSCAN target individual action spatiotemporal feature value extraction method based on the above-mentioned land, sea and air confrontation situation data is as follows:

[0084] STEP 1: Mark all historical behavior records of each target The initial state is True;

[0085] STEP 2: Randomly extract a behavior record , replace its mark to False: Draw a circle with the center point and traverse the behavior data set corresponding to the target of the behavior record All objects marked as True in ;

[0086] STEP 3: Determine the behavior record If Centered on The historical behavior frequency within the circle with radius , then mark the behavior record as the cluster center and create the corresponding cluster , take the target recording time in the cluster The average value is the standard time of the cluster , and then randomly select the next behavior record , go to STEP 2 until the target All historical behavior records Traverse all records (that is, mark all records as False) and go to STEP 4;

[0087] STEP 4: If the target If the total number of marked center points is 0, it means that the target has no fixed behavior pattern, and directly go to STEP 8: If the target If the total number of marked center points is not 0, go to STEP 5;

[0088] STEP 5: For any two center points in a cluster, if the time interval between their records is less than the EPS, the two points are connected. The connected center points and all the points in their neighborhood form a combined cluster. If a center point is not connected to any other center point, then the center point and all the points in its neighborhood form a cluster of its own. The historical behavior records that do not form a cluster are recorded as noise points, indicating the random behavior of the individual target.

[0089] STEP 6: Output individual goals All clusters generated

[0090] ;

[0091] STEP7: All clusters Inner key area With coordinate set Corresponding matching, we can get the spatiotemporal characteristics of individual behavior, where is the track segment number generated based on the target action key point set:

[0092] ;

[0093] Then integrate the output target The spatiotemporal characteristic matrix of individual behavior

[0094] ;

[0095] STEP 8: Enter the historical behavior record set for the next target ID and go to STEP 1. The algorithm ends when all target IDs have been traversed.

[0096] A prediction model for the spatiotemporal behavior characteristics of a target individual is used to analyze and predict the behavior patterns of an individual within a specific spatiotemporal range. This model primarily utilizes deep learning models (such as DNN networks and LSTM neural networks) to build the prediction model and trains it using historical data. After model training is complete, real-time data is fed into the model to predict the target individual's behavior patterns. In one embodiment, the construction of a prediction model based on the spatiotemporal behavior characteristics of a target individual includes:

[0097] (1) Coding network construction

[0098] The structure of the prediction network model mainly consists of the encoding layer - LSTM network layer - decoding layer, where the network structure of the encoding layer is as follows: Figure 4 As shown, it is mainly composed of a 2-layer DNN neural network. The input of the model is the feature vector of the above target:

[0099] ;

[0100] Output the encoded intermediate vector , as the input of the subsequent LSTM network layer.

[0101] (2) LSTM network layer construction

[0102] The LSTM network is a special recurrent neural network. Unlike general feedforward neural networks, the LSTM network can use the target's time series behavior data to analyze the input. Figure 5 middle, is the input information at time t, is the hidden vector of the network input at time t. We can see that the network recursively calls itself and passes the information at time t−1 to time t.

[0103] This network has a certain memory capacity and is suitable for processing time series data. When using this network, the feature vector given by the encoding network is used as The data are input into the network in sequence, and the network finally integrates a coding representation c of the current target state, which will serve as the input of the subsequent decoding layer network.

[0104] (3) Decoding layer network construction

[0105] refer to Figure 6 The decoding layer network consists of another fully connected network, whose output is the input of the LSTM network, and outputs the probability distribution of the next key area or key point of the target The goal of this network is mainly to generate a specific label probability distribution based on the encoding representation c of the current state of the target, and select the label with the largest probability as the prediction result during actual reasoning.

[0106] In one embodiment, the target individual spatiotemporal behavior feature prediction model inputs the previous feature vectors into the network model in chronological order each time it makes a prediction, and the model ultimately gives the probability distribution of the target's next key area or key point. , use the actual key points or key areas in the dataset during training One-Hot encoding representation of As labels, calculate the loss function:

[0107] ;

[0108] During training, the SGD optimizer is used to train on the target feature set to obtain the parameters of the model.

[0109] Compared with the existing target situation analysis method, the target action area clustering prediction method based on the time series network model proposed in the present invention has higher prediction accuracy and real-time performance. It can quickly identify and predict the target's action trend in massive data, thereby making more scientific and reasonable decisions.

[0110] As used herein, the word "preferred" is intended to serve as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, the use of the word "preferred" is intended to present concepts in a concrete manner. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or clear from the context, "X employs A or B" is intended to mean any of the naturally inclusive permutations. That is, if X employs A; X employs B; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing examples.

[0111] Moreover, although the present disclosure has been shown and described with respect to one or implementation, those skilled in the art will think of equivalent variations and modifications based on reading and understanding of this specification and the accompanying drawings. The present disclosure includes all such modifications and variations and is limited only by the scope of the appended claims. In particular, with respect to the various functions performed by the above-mentioned components (such as elements, etc.), the terms used to describe such components are intended to correspond to any component (unless otherwise indicated) that performs the specified function of the component (such as it is functionally equivalent), even if structurally different from the disclosed structure that performs the function in the exemplary implementation of the present disclosure shown herein. In addition, although the specific features of the present disclosure have been disclosed with respect to only one of several implementations, such features can be combined with one or other features of other implementations that can be desired and advantageous for a given or specific application. Moreover, insofar as the terms "including", "having", "containing" or their variations are used in specific embodiments or claims, such terms are intended to be included in a manner similar to the term "comprising".

[0112] The functional units in the embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or multiple or more units may be integrated into a single module. The aforementioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The aforementioned storage medium may be a read-only memory, a magnetic disk, or an optical disk, etc. The aforementioned devices or systems may execute the storage method in the corresponding method embodiment.

[0113] In summary, the above embodiment is one implementation method of the present invention, but the implementation method of the present invention is not limited to the described embodiment. Any other changes, modifications, substitutions, combinations, and simplifications that deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A target action area clustering prediction method based on a temporal network model, characterized in that: The following steps are involved: S1: Preprocessing of raw situation data, including data screening and data cleaning; S2: Build a target behavior profile based on business rules and specific situation data content to determine the target behavior characteristics; S3: Use the DBSCAN algorithm to mine the predicted target situation data and calculate the behavioral characteristics of all targets through clustering. The clustering results form a feature dataset about the target, which is used to train the subsequent neural network model. S4: Based on the number of features in the feature set, a target behavior prediction network model based on the LSTM structure is established, and the training loss function of the network model is determined; S5: Train the network model based on the SGD algorithm. After the training converges, export the model weights and network structure to a file for saving. S6: Import the trained model weights, apply them to real-time target data, and predict the target's action or area. Step S3 further includes: S31: Mark all historical behavior records of each target The initial state is True; S32: Randomly extract a behavior record , replace its mark to False: Draw a circle with the center point and traverse the behavior data set corresponding to the target of the behavior record All objects marked as True in ; S33: Determine the behavior record Attributes: If Centered on The frequency of historical behaviors within the circle with radius , then mark the behavior record as the cluster center and create the corresponding cluster , record the target time in the cluster The average value is taken as the standard time of the cluster , and then randomly select the next behavior record , turn to S32 until the target All historical behavior records All are traversed, that is, all behavior records are marked as False, and the process goes to S34; S34: If the target If the total number of marked center points is 0, it means that the target has no fixed behavior pattern and directly goes to S38. If the total number of marked center points is not 0, proceed to S35; S35: For any two center points in a cluster, if the time interval between their records is less than EPS, the two points are connected, and the connected center points and all the points in their neighborhood form a combined cluster; if a center point is not connected to any other center point, then the center point and all the points in its neighborhood form a cluster of its own: the historical behavior records that do not form a cluster are recorded as noise points, indicating the random behavior of individual targets; S36: Output individual goals All clusters generated: ； S37: All clusters Inner key area With coordinate set Corresponding matching, we can get the spatiotemporal characteristics of individual behavior, where is the track segment number generated based on the target action key point set: ； Then integrate the output target The spatiotemporal characteristic matrix of individual behavior; ； S38: Input the historical behavior record set of the next target ID and go to S31 until all target IDs are traversed and the algorithm ends.

2. The target action area clustering prediction method based on the temporal network model according to claim 1 is characterized in that: The original situation data is described mathematically as: ； In the above formula, Indicates the target No. The spatiotemporal characteristic value of individual action, Indicates the target No. The time period when the spatiotemporal characteristics of similar behaviors occur, Indicates the target No. The spatiotemporal characteristics of class behaviors correspond to the key area numbers, Indicates the target No. The spatiotemporal characteristics of class behaviors correspond to the frequency of historical behaviors. Respectively represent the target No. The spatiotemporal characteristics of the action-like objects correspond to the longitude and latitude coordinates of key area points. Indicates the target No. The spatiotemporal characteristics of the action-like objects correspond to the action route numbers; Target The longitude and latitude coordinates of the center or key points of the historical action key area .

3. The target action area clustering prediction method based on the temporal network model according to claim 1 is characterized in that: The data screening is carried out by comparing target intelligence information, the geographical location of key areas and relevant business rules, screening out all target situation data in all historical periods, eliminating behavioral data related to targets that are not of interest, retaining data records with target ID numbers, and eliminating data records without clear target IDs; The data cleaning cleans up the repeated fields in the behavior records to ensure the uniqueness of each behavior record; and all types of information in each behavior record remain complete, otherwise it will be directly deleted as abnormal data; at the same time, behavior information that is not related to the analysis of the spatiotemporal characteristics of individual target behaviors is deleted.

4. The target action area clustering prediction method based on the temporal network model according to claim 3 is characterized in that: The target behavior prediction network model based on the LSTM structure includes an encoding network, an LSTM network and a decoding network; The encoding network consists of a 2-layer DNN neural network, and the input of the model is the feature vector of the target: ； Output the encoded intermediate vector , as the input of the subsequent LSTM network layer; The LSTM network uses the target's time series behavior data to analyze the input; is the input information at time t, is the hidden vector of the network input at time t. The LSTM network recursively calls itself and passes the information at time t−1 to time t, taking the feature vector given by the encoding network as The LSTM network finally integrates a coding representation c of the current target state, which is used as the input of the subsequent decoding layer network. The decoding network consists of a fully connected network, whose output is the input of the LSTM network, and outputs the probability distribution of the next key area or key point of the target ,The decoding network generates a specific label probability distribution based on the encoding representation c of the current state of the target, and selects the label with the largest probability as the prediction result.

5. The target action area clustering prediction method based on the temporal network model according to claim 4 is characterized in that: The target behavior feature prediction model inputs the previous feature vectors into the network model in chronological order each time it makes a prediction, and the model finally gives the probability distribution of the next key area or key point of the target. , use the actual key points or key areas in the dataset during training One-Hot encoding representation of As labels, calculate the loss function: ； During training, the SGD optimizer is used to train on the target feature set to obtain the parameters of the model.

Images