CN114372620A

CN114372620A - Target person dynamic risk early warning method based on track prediction and related equipment

Info

Publication number: CN114372620A
Application number: CN202111629117.2A
Authority: CN
Inventors: 朵思惟; 余梓飞; 张程华; 于锋杰
Original assignee: Tianjin Huizhi Xingyuan Information Technology Co ltd
Current assignee: Tianjin Zhonghuan System Construction Co ltd
Priority date: 2021-12-28
Filing date: 2021-12-28
Publication date: 2022-04-19
Also published as: CN116128282A

Abstract

The application provides a target person dynamic risk early warning method based on track prediction and related equipment, wherein the method comprises the following steps: acquiring data information of a target person; dividing the data information into basic information and historical time sequence characteristic information; constructing a target person knowledge graph according to the basic information, and determining a target person characteristic value list based on the target person knowledge graph; inputting the historical time sequence characteristic information into a time sequence prediction model which is constructed in advance, and outputting predicted time sequence characteristic information through the time sequence prediction model; merging the target person characteristic value list and the predicted time sequence characteristic information to obtain a future time period characteristic value list; and obtaining the risk score of the target person through a pre-constructed risk studying and judging model based on the future time period characteristic value list. According to the method and the device, timeliness, accuracy and interpretability of the prediction early warning result are considered, and the problem of hysteresis of the traditional early warning technical scheme is solved.

Description

Target person dynamic risk early warning method based on track prediction and related equipment

Technical Field

The application relates to the technical field of machine learning, in particular to a target person dynamic risk early warning method based on track prediction and related equipment.

Background

With the rapid development of city intelligence and the rapidly growing city population, passive urban security solutions have become the past, and risk personnel trajectory tracking as an important ring of urban security faces new challenges.

At present, in the aspect of target personnel management and control, mass information data of risk personnel are analyzed mainly by means of fixed technical and combat laws or manual analysis, the potential risks are difficult to prejudge, risk behaviors occur before early warning, the early warning has certain hysteresis, and the risk cannot be prevented.

Therefore, a prediction and early warning method is needed to effectively early warn and control the potential risk of the target personnel.

Disclosure of Invention

In view of this, an object of the present application is to provide a target person dynamic risk early warning method based on trajectory prediction, which is characterized by comprising:

acquiring data information of a target person;

dividing the data information into basic information and historical time sequence characteristic information;

constructing a target person knowledge graph according to the basic information, and determining a target person characteristic value list based on the target person knowledge graph;

inputting the historical time sequence characteristic information into a time sequence prediction model which is constructed in advance, and outputting predicted time sequence characteristic information through the time sequence prediction model;

merging the target person characteristic value list and the predicted time sequence characteristic information to obtain a future time period characteristic value list;

and obtaining the risk score of the target person through a pre-constructed risk studying and judging model based on the future time period characteristic value list.

Based on the same inventive concept, the application also provides a target person dynamic risk early warning device based on track prediction, which comprises:

the acquisition module is configured to acquire data information of a target person;

a dividing module configured to divide the data information into basic information and historical timing characteristic information;

a construction module configured to construct a target person knowledge graph from the basic information, and determine a target person feature value list based on the target person knowledge graph;

the prediction module is configured to input the historical time sequence characteristic information into a time sequence prediction model which is constructed in advance, and output predicted time sequence characteristic information through the time sequence prediction model;

the merging module is configured to merge the target person characteristic value list and the predicted time sequence characteristic information to obtain a future time period characteristic value list;

a study and judgment module configured to obtain a risk score of the target person through a pre-constructed risk study and judgment model based on the future period feature value list

Based on the same inventive concept, the present application further provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable by the processor, wherein the processor implements the method as described above when executing the computer program.

Based on the same inventive concept, the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method as described above.

From the above, the target person dynamic risk early warning method based on trajectory prediction and the related device provided by the application predict the characteristics of the target person with time sequence attributes based on the deep learning Encoder-Decoder Encoder-Decoder framework, convert the target person future behavior prediction problem into the problem of 'known indefinite length input prediction indefinite length output', and apply the long-short term memory LSTM deep neural network to perform coding analysis on the indefinite length input and decode the input to generate indefinite length output. And predicting the change trend of the given characteristic of the target person in a future period of time based on the time sequence change rule of the given characteristic of the target person in a certain period of time, and giving early warning to the dynamic risk of the target person based on the prediction information. The method comprises the steps of combining dynamic information of target personnel with time sequence characteristics obtained through prediction and other basic information of the target personnel extracted from big data, taking each target personnel as a sample, generating labeled data of the target personnel of different types according to risk indexes of the target personnel of different types, constructing a dynamic risk early warning model of the target personnel based on a machine learning classification algorithm, training the dynamic risk early warning model of the target personnel by taking the labeled data as a training set, finally outputting the dynamic risk early warning model of the target personnel, outputting the model as risk scores of the target personnel, adjusting an early warning threshold value according to actual conditions, and finally giving out judgment on whether early warning is performed or not. The method and the device predict the characteristic that the target person has the time sequence attribute, predict the potential risk of the target person in the future period by combining the basic information of the target person, give consideration to timeliness, accuracy and interpretability of the prediction early warning result, and solve the problem of hysteresis of the traditional early warning technical scheme.

Drawings

In order to more clearly illustrate the technical solutions in the present application or the related art, the drawings needed to be used in the description of the embodiments or the related art will be briefly introduced below, and it is obvious that the drawings in the following description are only embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.

Fig. 1 is a schematic flowchart of a target person dynamic risk early warning method based on trajectory prediction according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a first knowledge-graph of an embodiment of the present application;

FIG. 3 is a schematic flow chart of a knowledge-graph completion according to an embodiment of the present application;

FIG. 4 is a schematic diagram illustrating a prediction flow of a time sequence prediction model according to an embodiment of the present application;

fig. 5 is a structural diagram of a target person dynamic risk early warning method device based on trajectory prediction according to an embodiment of the present application;

fig. 6 is a schematic diagram of a hardware structure of an electronic device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is further described in detail below with reference to the accompanying drawings in combination with specific embodiments.

It should be noted that technical terms or scientific terms used in the embodiments of the present application should have a general meaning as understood by those having ordinary skill in the art to which the present application belongs, unless otherwise defined. The use of "first," "second," and similar terms in the embodiments of the present application do not denote any order, quantity, or importance, but rather the terms are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that the element or item listed before the word covers the element or item listed after the word and its equivalents, but does not exclude other elements or items. The terms "connected" or "coupled" and the like are not restricted to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", and the like are used merely to indicate relative positional relationships, and when the absolute position of the object being described is changed, the relative positional relationships may also be changed accordingly.

As described in the background art, from the work history experience of the target person management and control, the behavior of the target person has certain periodicity and regularity, and with the inspiration, the transition from the traditional post-analysis studying and judging mode to the pre-prediction early warning mode is promoted by predicting the action track of the target person and fusing big data history information for feature analysis on the basis, so that the effective early warning and management and control on the potential risk of the target person are realized.

Embodiments of the present application are described in detail below with reference to the accompanying drawings.

The application provides a target person dynamic risk early warning method based on track prediction, and with reference to fig. 1, the method comprises the following steps:

and S101, acquiring data information of the target person.

Obtaining mass information resources related to target personnel, including basic information, historical information, daily activity information, vehicle access information, civil aviation information, railway information, passenger transport information, accommodation information, violation information, case event information and other multi-source heterogeneous data information.

And step S102, dividing the data information into basic information and historical time sequence characteristic information. The basic information in this embodiment mainly includes factors of individual and family situation dimensions. The historical timing characteristics information may have attributes of different timing dimensions, such as periodicity, continuous increase, continuous decrease, irregular change, and the like. By analyzing the historical time sequence characteristic information, the track of the target person in a period of time in the future can be predicted, when the periodic historical time sequence characteristic information changes, for example, the periodicity is damaged, the characteristic of the target person is abnormal, the target person can be continuously concerned, and possible risk behaviors are prevented in advance.

And S103, constructing a target person knowledge graph according to the basic information, and determining a target person feature value list based on the target person knowledge graph.

The construction of the knowledge graph refers to a process of extracting knowledge elements from original data by adopting a series of automatic or semi-automatic technical means from the original data and storing the knowledge elements into a knowledge base. The key technology of the knowledge graph is knowledge extraction, also called triple element extraction, and by the technology, knowledge elements such as entities, relations, attributes and the like can be extracted from data of some public semi-structured, unstructured and third-party structured databases. In this embodiment, a target person knowledge graph is constructed based on the basic information, the pre-labeled feature values of the target person with the missing value are complemented through the constructed target person knowledge graph, and the target person feature value list is obtained after completion.

And step S104, inputting the historical time sequence characteristic information into a time sequence prediction model which is constructed in advance, and outputting the predicted time sequence characteristic information through the time sequence prediction model.

Specifically, the time sequence prediction model is pre-trained, historical time sequence characteristic information of m days is input into the time sequence prediction model, predicted time sequence characteristic information of n days in the future can be output through the time sequence prediction model, predicted time sequence characteristic information of different days in the future can be obtained through training of different training sets, and the specific predicted time sequence model can be adjusted and trained according to actual conditions.

And S105, combining the target person characteristic value list and the predicted time sequence characteristic information to obtain a future time period characteristic value list.

And combining the predicted time sequence characteristic information obtained through prediction of the predicted time sequence model with a target person characteristic value list obtained based on the basic information to obtain a future time period characteristic value list, wherein the future time period characteristic value list is used as input information for risk research and judgment of the target person.

And S106, obtaining the risk score of the target person through a pre-constructed risk studying and judging model based on the future time period characteristic value list.

And inputting the characteristic value list of the future time period into a risk studying and judging model for calculation, outputting a risk score of the target person through the risk studying and judging model, wherein the risk score is 0 to represent low risk, the risk score is 1 to represent high risk, and special attention should be paid to the target person. The timeliness and the accuracy of the early warning of the target personnel are improved, and the problem of early warning lag is solved.

In some embodiments, the constructing a target person knowledge graph according to the basic information, and determining a target person feature value list based on the target person knowledge graph includes:

constructing a risk assessment index system according to the basic information; carrying out data annotation on the basic information according to the risk assessment index system to obtain an initial characteristic value list; determining a first entity relationship triple through entity extraction and relationship extraction based on the initial characteristic value list, and constructing an initial knowledge graph based on the first entity relationship triple; based on the initial knowledge graph, normalizing all entities in the initial knowledge graph through entity alignment to obtain a first knowledge graph; completing the first knowledge graph based on a preset abstract rule base to obtain the target person knowledge graph; and completing the initial characteristic value list based on the target person knowledge graph to obtain the target person characteristic value list.

Specifically, in this embodiment, according to the historical information of different types of target people and the suggestions of industry experts, a risk assessment index system of different types of target people is constructed, which covers the basic information of the target people, and the basic information mainly includes factors of the target individual and family condition dimensions and specific risk assessment indexes of different types of target people. The specific risk assessment index system is shown in table 1.

TABLE 1 Risk assessment index System

According to the constructed risk assessment index system of the target person, the data of the target person is labeled, specifically, as shown in table 2, the data form after labeling is shown, and because the collected data may have a situation that a part of index values are missing, missing value completion needs to be performed on the data as much as possible in the subsequent process.

TABLE 2 initial feature value List

For the initial feature value list in this embodiment, an initial knowledge graph is constructed through entity extraction and relationship extraction. For the table data, preliminary knowledge representation can be performed after data integration, and if text data exists, knowledge extraction needs to be performed on the text data, namely entity extraction, relationship extraction and attribute extraction are performed respectively, and then preliminary knowledge representation is performed.

The knowledge extraction technology is mainly divided into two parts, one part is entity extraction and named entity identification, and the other part is relationship extraction. The more detailed division divides element extraction into three parts, including attribute extraction in addition to the entity identification and relationship extraction described above. Since the attribute of an entity can be regarded as a special noun relationship between the entity and the attribute value, here we regard the attribute extraction as a special case of the relationship extraction, and the processing mode and algorithm are consistent with the relationship extraction. The entity types defined in this embodiment are shown in table 3.

TABLE 3 entity types

The defined relationship types mainly comprise the relationship of people, the modification relationship among various entities and the like.

The method adopts a deep learning bidirectional long-time memory network (Bi-LSTM) and a Conditional Random Field (CRF) model to learn the marked text sequence, constructs an algorithm model for named entity recognition, and extracts entity elements in the evaluation index system.

The LSTM is an algorithm model for processing the serialized data, the algorithm solves the problem of dependence of the recurrent neural network on long-distance sequence information, can well solve the problems of gradient disappearance, gradient explosion and the like of the recurrent neural network, and has good effect when applied to an entity extraction task. The CRF is an undirected graph probability model for sequence data labeling, can give a conditional probability distribution of another group of output sequences according to a given group of input sequences, maximizes the probability of a target labeling sequence, and realizes sequence labeling on data to be labeled.

The method adopts a deep learning bidirectional long-time memory network conditional random field (BilSTM-CRF) to extract the entities. The model obtains word vector mapping of each word through a Look-up layer of the first layer, and a Dropout layer is arranged to prevent overfitting. The second layer is a bidirectional LSTM (BilsTM) layer, the word vectors enter the BilsTM layer, the score probability of each word corresponding to each label is output through the information of the learning context, and the bidirectional structure can be adopted to effectively find the structural relationship between the front and the back of the text. And finally, taking the output of the BilSTM layer as the input of the CRF layer, and obtaining a final prediction result by learning the sequence dependence information among the labels. The CRF layer can correct the output of the BilSTM layer through learning the transition probability among the labels in the data set, thereby ensuring the rationality of the predicted labels. In the BilSTM-CRF algorithm, a BilSTM layer learns the context information of sequences, a CRF layer learns the dependency information among labels, and the algorithm can finally realize the accurate extraction of the entity elements such as people, places, organizations and the like through the combination of the two.

After the entities are extracted, the relationships between the entity pairs such as "person-person", "person-thing" and the like need to be further extracted. The method and the device perform relation classification based on the Chinese pre-training language model Bert-Chinese, so that the extraction of the relation is realized. The following 12 types of relationships are common in target person data: parents, couples, teachers and students, brothers and sisters, colleagues, collaborations, lovers, grandsons, friends, relatives, juniors and juniors, and the like. And finally outputting an entity-relation-entity triple set serving as an initial knowledge graph through entity extraction and relation extraction.

Based on the initial knowledge graph obtained in the previous step, in this step, entity alignment needs to be performed on entity nodes which are different in calling but point to the same entity in the extracted entity-relationship-entity triple set. For example, "three of plum" and "one of plum" and "duan-wife-wangwu" in "three of plum-couple-wangwu" and "duan-duan" both refer to the entity "three of plum", and these different expressions need to be normalized by entity alignment, and finally, a complete first knowledge graph is constructed, as shown in fig. 2. Entity alignment is the process of determining whether two entities point to the same object in the real world. The purpose of entity alignment is to find entities with different entity names but representing the same thing in the real world, merge the entities, identify the entity with a unique identifier, and finally add the entity to the corresponding knowledge graph. The entity alignment method based on similarity propagation is adopted to regard the entity alignment problem as an optimization problem of a global matching scoring objective function for modeling. The problem belongs to a binary classification problem, and an approximate solution can be obtained through a greedy optimization algorithm.

The embodiment completes the first knowledge graph based on the hybrid reasoning algorithm, so as to obtain more complete target person information, and prepare for subsequently completing missing values in the original characteristic value list of the target person. The mixed inference algorithm firstly carries out embedded expression on the first knowledge graph, the nodes are embedded into vectors, and the edge relations are embedded into matrixes. And constructing an initial instance rule base of the knowledge graph based on the first knowledge graph and a predefined abstract rule base, and screening out reasonable instance rules from the initial instance rule base based on an embedded matrix of edge relations to serve as a final reasonable instance rule base. And reasoning to obtain a new triple according to the triple reasoning rule corresponding to each example rule based on the obtained reasonable example rule base, thereby completing the completion of the first knowledge graph and obtaining the target person knowledge graph. And completing the initial characteristic value list based on the obtained target person knowledge graph to obtain the target person characteristic value list.

In some embodiments, the complementing the first knowledge-graph based on a preset abstract rule base to obtain the target person knowledge-graph includes:

traversing all second entity relationship triples in the first knowledge graph, and combining all example rules corresponding to the second entity relationship triples which meet the abstract rule base to serve as an initial example rule base; constructing a negative example entity relationship triple according to the second entity relationship triple; inputting the second entity relationship triple and the negative case entity relationship triple into a pre-constructed knowledge map embedding model, and outputting vector embedding representation of the entity and matrix embedding representation of the relationship; calculating a confidence score for each instance rule in the initial instance rule base based on the vector embedded representation of the entity and the matrix embedded representation of the relationship; merging all the example rules with the confidence scores exceeding a preset confidence threshold value to serve as a reasonable example rule base; reasoning according to the reasonable instance rule base to obtain a complementary entity relationship triple based on the second entity relationship triple; and supplementing the supplemented entity relationship triples into the first knowledge graph to obtain the target person knowledge graph.

The embodiment completes the knowledge graph based on knowledge graph mixed reasoning, further perfects the information of the target personnel, and constructs a more comprehensive labeled data sample. Generally speaking, the target person information may have some intentional or unintentional imperceptibility, and the imperceptibility causes information loss, and the missing information is difficult to be directly inferred in a manual manner in a large amount of sample information. The covertness reflects the path relation corresponding to the graph structure in the knowledge graph, and the method based on mixed reasoning in knowledge reasoning completes the entity and entity-side relation of hidden relation variables and characteristics in the prior knowledge graph and then further completes the missing value of the initial characteristic value list based on the completed knowledge graph.

Specifically, the first knowledge graph is embedded and represented. The embedded model can be obtained by training common knowledge graph embedded models ANALOGY, RESCAL and the like and minimizing the following loss function

Where L is a loss function, n is the total number of input triples, σ (-) is a sigmoid function,

representing the embedded representation of subject s and object o in triples,

matrix embedding representing the relation r in triplets, m being the embedding dimension, l_sroTags corresponding to the embedded triplets.

The input to the embedded learning is a set of triples and their corresponding tags:

I＝{((s,r,o),l_sro)|(s,r,o)∈G∪G_neg}

the values of the tags are defined as follows:

the triples in the set include the "entity-relationship-entity" triples (s, r, o) ∈ G and the constructed negative triples (s, r, o) ∈ G in the first knowledge-graph_neg. The triple negative example can be obtained by replacing s and o of the triple in the first knowledge graph with any entity in the first knowledge graph or replacing the relation r with any relation in the relation of the first knowledge graph. Taking FIG. 3 as an example, (target person 1, residence, Beijing) ∈ G belongs to the triplet in the first knowledge base, (target person 2, couple, Beijing) ∈ G_negBelonging to the negative examples of the triples.

As shown in fig. 3, traversing all second entity relationship triples in the first knowledge-graph, and merging all instance rules corresponding to the second entity relationship triples that satisfy the abstract rule base to serve as an initial instance rule base. The abstract rule base is shown in table 4,

TABLE 4 Abstract rule base

For a given abstraction rule as defined in table 4 above, all edge relationships in the graph are traversed and, if there is an instance that meets the abstraction rule, added to the initial instance rule base. Taking the abstract rule 'symmetric attribute' as an example, the triplet inference satisfying the abstract rule instance in the map is expressed as:

(target person 1, couple, target person 2) → (target person 2, couple, target person 1)

The corresponding example rule is as follows: symmetric attribute (couple), put the instance into the initial instance rule base.

Some of the example rules in the initial example rule base are listed in Table 5

TABLE 5 initial example rule base

Initial instance rule base
	Symmetry attribute (couple)
Symmetry property (friend)
	Delivery attribute (friend)
Genus of equivalenceSex (birth date)
	Reversible attribute (child, parent)
Reversible attribute (daughter, mother)
	Reversible attribute (child, father)
The attribute chain includes ((couple, place of residence)
	……

Since some example rules in the obtained initial example rule base do not conform to logic, for example, "friend" does not necessarily satisfy the transfer attribute in some cases, we need to further filter the initial example rule base to obtain a more general and reasonable example rule base. Given the embedded representation of all relationships in the first knowledge-graph and the initial instance rule base described above, in order to screen out a reasonable instance rule base, we need to give a confidence score for each instance in the initial instance rule base for the abstract rule to which it belongs. The confidence score is calculated as follows

Wherein |_FIs a Frobenius norm and is used for measuring the similarity degree between two matrixes

And

respectively represent equivalent symbols in column 3 of Table 4

The right side is equal-sign two sides. For example, for "symmetric properties", the matrix

Matrix array

Confidence score s_a∈[0,1]Selecting 0.9 as threshold in practical application, and finally making all confidence scores meet s_aAnd the reasonable example rule base consisting of the example rules of more than or equal to 0.9 is used as output, as shown in the table 6.

TABLE 6 reasonable example rule base

And reasoning to obtain a complementary entity relationship triple according to the triple reasoning rule corresponding to each instance rule based on the obtained reasonable instance rule base. For example, the triple inference corresponding to the example rule "((couple, residence)" is represented as

(x₀,r₁,x₁),(x₁,r₂,x₂)→(x₀,r₂,x₂)

Thus, from the triplets in the first knowledge-graph (target person 2, couple, target person 1), (target person 1, place of residence, Beijing), a complementary entity relationship triplet (target person 2, place of residence, Beijing) may be inferred.

The target person knowledge-graph (including dashed lines) in fig. 3 is a complementary knowledge-graph to the first knowledge-graph (not including dashed lines) after the reasoning. In the original initial feature value list, the value of the "living environment" of the target person 2 is missing, and the "living environment" of the target person can be inferred through the relationship between the target person and his wife in the knowledge map and the public security environment of the living place and the living place of the wife (the target person 1), so that more comprehensive target person information can be obtained.

Based on the completed target person knowledge graph, mapping the corresponding completed value back to the initial characteristic value list, completing the missing value, quantizing part of descriptive characteristics, converting the quantized values into numerical values, taking the economic condition as an example, "good-2, medium-1, poor-0", and finally outputting a complete target person characteristic value list as shown in table 7.

TABLE 7 target person feature value List

In some embodiments, the time-series prediction model includes an encoder and a decoder, both of which are recurrent neural networks, the inputting the historical time-series characteristic information into a pre-constructed time-series prediction model, and outputting predicted time-series characteristic information through the time-series prediction model includes:

inputting the historical timing characteristic information into the encoder, outputting a state vector via the encoder; inputting the state vector into the decoder, outputting the prediction timing feature information via the decoder.

Specifically, in the present embodiment, an Encoder-Decoder framework is employed, since both the input and output are of indefinite length sequences due to uncertainty in the number of visits by the target person within a given time period. The encoder and decoder are two recurrent neural networks corresponding to the input sequence and the output sequence, respectively. In the embodiment, the RNN network, i.e., the encor, for analyzing and processing the input with an indefinite length, and the network, i.e., the decor, for generating the output with an indefinite length are used, and both of them form a network framework, i.e., the encor-decor. And taking the example time sequence characteristic information as the input of the encoder to obtain a state vector, and inputting the state vector into a decoder to obtain the predicted time sequence characteristic information.

In some embodiments, the predicting the time-series feature information includes a plurality of predicted trace points, the inputting the state vector into the decoder, and outputting the predicted time-series feature information via the decoder include:

encoding a starting character, merging the encoded starting character and the state vector to serve as a new state vector to be input into an input layer of the decoder, and outputting the new state vector to obtain a first predicted track point through the decoder;

except for the first predicted track point, each of the remaining predicted track points is obtained by the following operations: coding the predicted track point which is one track point before the predicted track point, merging the coded predicted track point and a state vector used for predicting the previous track point to serve as a state vector of the predicted track point, inputting the state vector into an input layer of a decoder, and outputting the state vector to obtain the predicted track point through the decoder;

and in response to the fact that the end character is determined to be detected, stopping outputting the predicted track points, and sequentially combining all the output predicted track points to serve as the predicted time sequence characteristic information.

Specifically, as shown in fig. 4, in this embodiment, the historical time series characteristic information is [ home, market, home, school, home ], and is input as an input sequence into an embedding layer to be embedded and represented, an encoder encodes the historical time series characteristic information to obtain a hidden vector as a state vector, and simultaneously, a START character < START > is embedded and represented through the embedding layer and is input into a decoder together with the state vector, a position of a next track point is predicted through an argmax function in a Softmax layer, and a preset position dictionary is used to retrieve and match a corresponding track point, where the predicted track point is [ cafe ], and [ cafe ] is placed into the predicted time series characteristic information. Then, coding is carried out on the (coffee hall) and the coding and state vectors of the starting characters are input into a decoder again for prediction, the position of the next track point is predicted through an argmax function, corresponding track points are searched and matched through a preset position dictionary, the next predicted track point (square) is obtained, the like is carried out, the track point (square) is coded and then input into the decoder together with the coding vectors and the state vectors of all the previous predicted track points for prediction, the predicted track point (home) is obtained, the ending character < END > is detected, prediction is finished, and the prediction time sequence characteristic information (coffee hall, square and home) is obtained.

In some embodiments, the risk study model is pre-trained by a limit gradient boosting XGBoost algorithm.

Taking each person as a sample, aiming at different types of target persons, taking corresponding risk index data as the characteristics of the sample, taking 70% of the total samples as a test set and 30% of the total samples as a verification set to be input into a regression model, and training the model by applying an extreme Gradient boosting XGboost (extreme Gradient boosting) algorithm. Besides high precision, the algorithm uses C + + in the back of the whole model, so the training speed is very fast, and the model is also adjusted a lot on the algorithm, such as highly utilizing CPU to perform multi-core parallel operation. The XGBoost is essentially a gradient Boosting Decision tree GBDT (gradient Boosting Decision tree) in a machine learning algorithm, but exerts speed and efficiency extremely on the basis of the algorithm, and can be regarded as an engineering implementation of the GBDT algorithm. In the training process, the disturbance range of the tuning parameter of the XGboost algorithm is as follows: the learning rate is 0.1-0.3, the maximum depth of the tree is 5-10, the sample sampling ratio is 0.7-l, the sample attribute sampling ratio is 0.7-1, the iteration frequency is 100-1000, and the regularization term weight is 5-10. The trained XGboost model is finally output as a risk study and judgment model.

It should be noted that the method of the embodiment of the present application may be executed by a single device, such as a computer or a server. The method of the embodiment can also be applied to a distributed scene and completed by the mutual cooperation of a plurality of devices. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiment, and the multiple devices interact with each other to complete the method.

It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments described above and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

Based on the same inventive concept, corresponding to the method of any embodiment, the application also provides a target person dynamic risk early warning device based on track prediction.

Referring to fig. 5, the trajectory prediction-based target person dynamic risk early warning apparatus includes:

an obtaining module 501 configured to obtain data information of a target person;

a dividing module 502 configured to divide the data information into basic information and historical timing characteristic information;

a construction module 503 configured to construct a target person knowledge graph according to the basic information, and determine a target person feature value list based on the target person knowledge graph;

a prediction module 504 configured to input the historical timing characteristic information into a pre-constructed timing prediction model, and output predicted timing characteristic information through the timing prediction model;

a merging module 505 configured to merge the target person feature value list and the predicted time series feature information to obtain a future time period feature value list;

a judging module 506 configured to obtain a risk score of the target person through a pre-constructed risk judging model based on the future time period feature value list.

For convenience of description, the above devices are described as being divided into various modules by functions, and are described separately. Of course, the functionality of the various modules may be implemented in the same one or more software and/or hardware implementations as the present application.

The device of the above embodiment is used to implement the corresponding target person dynamic risk early warning method based on trajectory prediction in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which are not described herein again.

Based on the same inventive concept, corresponding to the method of any embodiment described above, the present application further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the target person dynamic risk early warning method based on trajectory prediction described in any embodiment described above is implemented.

Fig. 6 is a schematic diagram illustrating a more specific hardware structure of an electronic device according to this embodiment, where the electronic device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein the processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 are communicatively coupled to each other within the device via bus 1050.

The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits, and is configured to execute related programs to implement the technical solutions provided in the embodiments of the present disclosure.

The Memory 1020 may be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, or the like. The memory 1020 may store an operating system and other application programs, and when the technical solution provided by the embodiments of the present specification is implemented by software or firmware, the relevant program codes are stored in the memory 1020 and called to be executed by the processor 1010.

The input/output interface 1030 is used for connecting an input/output module to input and output information. The i/o module may be configured as a component in a device (not shown) or may be external to the device to provide a corresponding function. The input devices may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices may include a display, a speaker, a vibrator, an indicator light, etc.

The communication interface 1040 is used for connecting a communication module (not shown in the drawings) to implement communication interaction between the present apparatus and other apparatuses. The communication module can realize communication in a wired mode (such as USB, network cable and the like) and also can realize communication in a wireless mode (such as mobile network, WIFI, Bluetooth and the like).

Bus 1050 includes a path that transfers information between various components of the device, such as processor 1010, memory 1020, input/output interface 1030, and communication interface 1040.

It should be noted that although the above-mentioned device only shows the processor 1010, the memory 1020, the input/output interface 1030, the communication interface 1040 and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, those skilled in the art will appreciate that the above-described apparatus may also include only those components necessary to implement the embodiments of the present description, and not necessarily all of the components shown in the figures.

The electronic device of the above embodiment is used to implement the corresponding target person dynamic risk early warning method based on trajectory prediction in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.

Based on the same inventive concept, corresponding to any of the above embodiments, the present application further provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the trajectory prediction based target person dynamic risk early warning method according to any of the above embodiments.

Computer-readable media of the present embodiments, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the target person dynamic risk early warning method based on trajectory prediction according to any of the above embodiments, and have the beneficial effects of corresponding method embodiments, which are not described herein again.

Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, is limited to these examples; within the context of the present application, features from the above embodiments or from different embodiments may also be combined, steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of brevity.

In addition, well-known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown in the provided figures for simplicity of illustration and discussion, and so as not to obscure the embodiments of the application. Furthermore, devices may be shown in block diagram form in order to avoid obscuring embodiments of the application, and this also takes into account the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform within which the embodiments of the application are to be implemented (i.e., specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the application, it should be apparent to one skilled in the art that the embodiments of the application can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative instead of restrictive.

While the present application has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic ram (dram)) may use the discussed embodiments.

The present embodiments are intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Therefore, any omissions, modifications, substitutions, improvements, and the like that may be made without departing from the spirit and principles of the embodiments of the present application are intended to be included within the scope of the present application.

Claims

1. A target person dynamic risk early warning method based on track prediction is characterized by comprising the following steps:

acquiring data information of a target person;

2. The method of claim 1, wherein constructing a target person knowledge graph from the base information, and determining a target person feature value list based on the target person knowledge graph comprises:

constructing a risk assessment index system according to the basic information;

carrying out data annotation on the basic information according to the risk assessment index system to obtain an initial characteristic value list;

determining a first entity relationship triple through entity extraction and relationship extraction based on the initial characteristic value list, and constructing an initial knowledge graph based on the first entity relationship triple;

based on the initial knowledge graph, normalizing all entities in the initial knowledge graph through entity alignment to obtain a first knowledge graph;

completing the first knowledge graph based on a preset abstract rule base to obtain the target person knowledge graph;

and completing the initial characteristic value list based on the target person knowledge graph to obtain the target person characteristic value list.

3. The method of claim 2, wherein determining a first entity-relationship triple by entity extraction and relationship extraction based on the initial feature value list comprises:

and performing entity extraction on the initial characteristic value list through a bidirectional long-time memory network and a conditional random field, and performing relationship extraction on the initial characteristic value list through a pre-training model Bert to determine the first entity relationship triple.

4. The method of claim 2, wherein the complementing the first knowledge-graph based on a preset abstract rule base to obtain the target person knowledge-graph comprises:

traversing all second entity relationship triples in the first knowledge graph, and combining all example rules corresponding to the second entity relationship triples which meet the abstract rule base to serve as an initial example rule base;

constructing a negative example entity relationship triple according to the second entity relationship triple;

inputting the second entity relationship triple and the negative case entity relationship triple into a pre-constructed knowledge map embedding model, and outputting vector embedding representation of the entity and matrix embedding representation of the relationship;

calculating a confidence score for each instance rule in the initial instance rule base based on the vector embedded representation of the entity and the matrix embedded representation of the relationship;

merging all the example rules with the confidence scores exceeding a preset confidence threshold value to serve as a reasonable example rule base;

reasoning according to the reasonable instance rule base to obtain a complementary entity relationship triple based on the second entity relationship triple;

and supplementing the supplemented entity relationship triples into the first knowledge graph to obtain the target person knowledge graph.

5. The method of claim 4, wherein constructing a negative instance entity relationship triplet from the second entity relationship triplet comprises:

replacing an entity in the second entity-relationship triplet with any entity in the first knowledge-graph,

and/or replacing the relationship in the second entity relationship triple with any relationship in the first knowledge-graph to obtain the negative example entity relationship triple.

6. The method of claim 1, wherein the temporal prediction model comprises an encoder and a decoder, both of which are recurrent neural networks,

the inputting the historical time sequence characteristic information into a time sequence prediction model which is constructed in advance, and outputting the predicted time sequence characteristic information through the time sequence prediction model comprises the following steps:

inputting the historical timing characteristic information into the encoder, outputting a state vector via the encoder;

inputting the state vector into the decoder, outputting the prediction timing feature information via the decoder.

7. The method of claim 6, wherein the predicted temporal feature information includes a plurality of predicted trace points,

the inputting the first state vector into the decoder, outputting the prediction timing feature information via the decoder, comprising:

except for the first predicted track point, each of the remaining predicted track points is obtained by the following operations:

coding the predicted track point which is one track point before the predicted track point, merging the coded predicted track point and a state vector used for predicting the previous track point to serve as a state vector of the predicted track point, inputting the state vector into an input layer of a decoder, and outputting the state vector to obtain the predicted track point through the decoder;

8. The method of claim 1, wherein the risk study model is pre-trained by a extreme gradient boost (XGboost) algorithm.

9. The utility model provides a target personnel developments risk early warning device based on track prediction which characterized in that includes:

and the studying and judging module is configured to obtain the risk score of the target person through a pre-constructed risk studying and judging model based on the future time period characteristic value list.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 8 when executing the program.