CN111898836A - Crime space-time prediction method and system - Google Patents
Crime space-time prediction method and system Download PDFInfo
- Publication number
- CN111898836A CN111898836A CN202010828611.0A CN202010828611A CN111898836A CN 111898836 A CN111898836 A CN 111898836A CN 202010828611 A CN202010828611 A CN 202010828611A CN 111898836 A CN111898836 A CN 111898836A
- Authority
- CN
- China
- Prior art keywords
- sequence
- date
- grid
- crime
- weather
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 239000013598 vector Substances 0.000 claims abstract description 256
- 239000011159 matrix material Substances 0.000 claims abstract description 149
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 17
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 238000003058 natural language processing Methods 0.000 claims abstract description 12
- 230000007246 mechanism Effects 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 12
- 238000011161 development Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 230000009545 invasion Effects 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 230000002265 prevention Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000011840 criminal investigation Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000019580 granularity Nutrition 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Primary Health Care (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a crime space-time prediction method and a system, wherein a map of a region to be predicted is subjected to grid division, a multi-granularity convolutional neural network data processing model is utilized to obtain a crime case number sequence, a weather sequence, a date sequence and a region sequence of each grid in a preset historical period, then a GloVe natural language processing method is utilized to express each sequence as a vector matrix, all the vector matrices are spliced into a multi-dimensional characteristic vector matrix of each grid according to dates, a weighted multi-dimensional characteristic vector matrix of each grid is determined based on a dynamic fusion algorithm model of a self-attention mechanism, a multi-window encoder is utilized to carry out information encoding on the weighted multi-dimensional characteristic vector matrix, the obtained encoding characteristics captured by windows with different lengths are used as a training set training classifier, and finally the trained classifier is utilized to predict the crime case occurrence number of each grid on a target date, the method realizes the efficient development of space-time analysis and prediction on crimes.
Description
Technical Field
The invention relates to the field of intelligent crime prediction, in particular to a crime space-time prediction method and a system.
Background
Although the artificial intelligence technology has made great progress in the crime field, due to the limitation of the difficulty in obtaining crime data, many achievements still only stay in the research level, cannot effectively meet actual combat, cannot serve actual combat, and the current crime related data are more and still stay in the query and retrieval application stage, so that the deep analysis, study and judgment and deep application of the data are very weak.
Disclosure of Invention
The invention aims to provide a method and a system for predicting a crime space-time to efficiently carry out space-time analysis and prediction on a crime.
In order to achieve the purpose, the invention provides the following scheme:
a criminal spatiotemporal prediction method, the prediction method comprising:
carrying out grid division on a map of an area to be predicted, and acquiring the time of criminal case occurrence of each grid in a preset historical period;
determining the number of crime cases of each grid in a preset historical period according to the occurrence time of the crime cases;
respectively filling the number of crime cases, the weather and the date of each day of each grid in a preset historical period and the region to which each grid belongs into sequences with the same length of days by taking the day as a unit by utilizing a multi-granularity convolutional neural network data processing model to obtain the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in the preset historical period;
respectively representing the crime case number sequence, the weather sequence, the date sequence and the region sequence as vector matrixes with the same dimensionality by using a GloVe natural language processing method, and splicing all the vector matrixes into a multi-dimensional characteristic vector matrix of each grid according to dates;
determining a weighted multidimensional characteristic vector matrix of each grid by utilizing a dynamic fusion algorithm model based on a self-attention mechanism according to the multidimensional characteristic vector matrix of each grid;
adopting a multi-window encoder to perform information encoding on the weighted multi-dimensional eigenvector matrix to obtain encoding characteristics captured by windows with different lengths;
training a classifier by taking the code characteristics captured by the windows with different lengths as a training set to obtain a trained classifier;
inputting the target date, the weather of the target date and the region to which each grid belongs into the trained classifier, and predicting the number of crime cases of each grid of the region to be predicted on the target date.
Optionally, the determining, according to the time of the crime case occurrence, the number of crime cases per day in a preset historical period for each grid, and then further comprising:
determining median X of all crime cases in each grid in preset historical periodmedium；
Calculating the absolute deviation value between the number of the criminal cases of each grid in a preset historical period and the median;
determining median MAD of all absolute deviation values;
according to median X of the number of the crime casesmediumAnd the median MAD of the absolute deviation value, and determining the range of the number of the crime cases as [ Xmedium-n×MAD,Xmedium+n×MAD]；
The number of crime cases is not in [ X ]medium-n×MAD,Xmedium+n×MAD]The value of the number of crime cases of the grids in the range is set to be 0;
where n is a constant scaling factor.
Optionally, the method includes, by using a multi-granularity convolutional neural network data processing model, respectively filling the number of crime cases, the weather, the date of each day, and the region to which each grid belongs in a preset historical period with a unit of day into sequences of the same length of days, to obtain the number sequence, the weather sequence, the date sequence, and the region sequence of crime cases of each grid in the preset historical period, and specifically includes:
expressing the number of crime cases per day in a preset historical period as an initial sequence of the number of cases according to dates;
representing the weather of each day in a preset historical period as an initial weather sequence according to the date;
representing the dates of each day in a preset historical period as an initial date sequence according to the dates;
representing the region to which each grid belongs in a preset historical period as a region initial sequence in a day unit according to the date;
and filling the case number initial sequence, the weather initial sequence, the date initial sequence and the region initial sequence into a sequence with a preset number of days by using a multi-granularity convolutional neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in a preset historical period.
Optionally, the step of using a GloVe natural language processing method to respectively represent the crime case number sequence, the weather sequence, the date sequence and the region sequence as vector matrices of the same dimension, and splice all the vector matrices into a multidimensional feature vector matrix of each grid according to the date specifically includes:
converting the number of crime cases in the crime case number sequence, the weather in the weather sequence, the date in the date sequence and the region in the region sequence into vectors with the same dimensionality respectively by using a trained word vector representation model based on a GloVe method;
all vectors in the crime case number sequence form a crime case number vector matrix, all vectors in the weather sequence form a weather vector matrix, all vectors in the date sequence form a date vector matrix, and all vectors in the region sequence form a region vector matrix;
and splicing the number vector matrix, the weather vector matrix, the number vector of the same date in the date vector matrix and the region vector matrix, the weather vector, the date vector matrix and the region vector to form the multi-dimensional characteristic vector matrix of each grid.
Optionally, the determining, according to the multidimensional feature vector matrix of each grid, a weighted multidimensional feature vector matrix of each grid by using a dynamic fusion algorithm model based on a self-attention mechanism specifically includes:
taking a vector corresponding to each date in the multi-dimensional feature vector matrix as an element of the multi-dimensional feature vector matrix, and assigning each element with a key value;
respectively calculating the similarity between the key value of the ith element and the key value of each element in the multi-dimensional feature vector;
normalizing all the similarity by utilizing softmax to obtain a normalized value of the similarity;
according to the formula xi′＝k1x1+k2x2+…+knxnObtaining the weighting vector x of the ith elementi′；
All the weighted vectors form a weighted multidimensional characteristic vector matrix of each grid;
wherein k is1Normalized value, k, of the similarity of the key value of the ith element to the key value of the 1 st element in the multi-dimensional feature vector2Normalized value, k, of the similarity of the key value of the ith element to the key value of the 2 nd element in the multi-dimensional feature vectornNormalized value, x, of similarity of key value of ith element and key value of nth element in multi-dimensional feature vector1、x2、xnRespectively a first element, a second element and an nth element of the multidimensional characteristic vector matrix.
Optionally, the training of the classifier by using the coding features of the size of the multiple windows as a training set to obtain the trained classifier specifically includes:
updating the parameters of the classifier by using a random gradient descent algorithm, and taking the parameter with the minimum loss function value as the optimal parameter of the classifier to obtain the trained classifier; the loss function is H (p, q) ═ Σ p (x) logq (x), where H (p, q) is the loss function, p (x) is the probability distribution of the number of real crime cases on the history date, and q (x) is the probability distribution of the number of predicted crime cases on the history date.
A criminal spatiotemporal prediction system, the prediction system comprising:
the criminal case occurrence time acquisition module is used for dividing grids of a map of an area to be predicted and acquiring criminal case occurrence time of each grid in a preset historical period;
the number determining module of the crime cases is used for determining the number of the crime cases of each grid in a preset historical period according to the occurrence time of the crime cases;
the sequence obtaining module is used for respectively filling the number of crime cases, the weather and the date of each day of each grid in a preset historical period and the region to which each grid belongs into sequences with the same length of days by taking the day as a unit by utilizing a multi-granularity convolutional neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in the preset historical period;
the multi-dimensional characteristic vector matrix determining module is used for respectively representing the crime case number sequence, the weather sequence, the date sequence and the region sequence as vector matrixes with the same dimension by using a GloVe natural language processing method, and splicing all the vector matrixes into a multi-dimensional characteristic vector matrix of each grid according to the dates;
the weighted multi-dimensional eigenvector matrix determining module is used for determining the weighted multi-dimensional eigenvector matrix of each grid by utilizing a dynamic fusion algorithm model based on a self-attention mechanism according to the multi-dimensional eigenvector matrix of each grid;
the coding feature capturing module is used for carrying out information coding on the weighted multidimensional feature vector matrix by adopting a multi-window coder to obtain coding features captured by windows with different lengths;
the trained classifier obtaining module is used for taking the code characteristics captured by the windows with different lengths as a training set to train the classifier so as to obtain a trained classifier;
and the crime case occurrence number prediction module is used for inputting the target date, the weather of the target date and the region to which each grid belongs into the trained classifier and predicting the crime case occurrence number of each grid of the region to be predicted on the target date.
Optionally, the sequence obtaining module specifically includes:
the case number initial sequence submodule is used for representing the number of crime cases per day in a preset historical period as a case number initial sequence according to dates;
the weather initial sequence submodule is used for representing the weather of each day in a preset historical period as a weather initial sequence according to the date;
the date initial sequence submodule is used for expressing the dates of each day in the preset historical period as a date initial sequence according to the dates;
the region initial sequence submodule is used for expressing the region to which each grid belongs in a preset historical period as a region initial sequence in a day unit according to the date;
and the sequence determining submodule is used for filling the case number initial sequence, the weather initial sequence, the date initial sequence and the region initial sequence into a sequence with a preset number of days by using a multi-granularity convolutional neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in a preset historical period.
Optionally, the multidimensional feature vector matrix determining module specifically includes:
the vector conversion submodule is used for converting the number of crime cases in the crime case number sequence, the weather in the weather sequence, the date in the date sequence and the region in the region sequence into vectors with the same dimension respectively by using a trained word vector representation model based on a GloVe method;
the vector matrix forming submodule is used for forming a crime case number vector matrix by all vectors in the crime case number sequence, forming a weather vector matrix by all vectors in the weather sequence, forming a date vector matrix by all vectors in the date sequence and forming a region vector matrix by all vectors in the region sequence;
and the multi-dimensional eigenvector matrix forming submodule is used for splicing the number vector matrix of the crime case, the weather vector matrix, the number vector of the date vector matrix and the region vector matrix on the same date, the weather vector, the date vector matrix and the region vector to form the multi-dimensional eigenvector matrix of each grid.
Optionally, the weighted multidimensional feature vector matrix determining module specifically includes:
the key value assignment submodule is used for taking a vector corresponding to each date in the multi-dimensional characteristic vector matrix as an element of the multi-dimensional characteristic vector matrix and assigning each element with a key value;
the similarity calculation operator module is used for respectively calculating the similarity between the key value of the ith element and the key value of each element in the multi-dimensional feature vector;
the similarity normalization value acquisition submodule is used for normalizing all the similarities by utilizing softmax to obtain a similarity normalization value;
obtaining submodule of weighted vector according to formula x'i＝k1x1+k2x2+…+knxnObtaining a weighting vector x 'of the ith element'i；
A weighted multidimensional characteristic vector matrix forming submodule for forming a weighted multidimensional characteristic vector matrix of each grid by all weighted vectors;
wherein k is1Normalized value, k, of the similarity of the key value of the ith element to the key value of the 1 st element in the multi-dimensional feature vector2Normalized value, k, of the similarity of the key value of the ith element to the key value of the 2 nd element in the multi-dimensional feature vectornNormalized value, x, of similarity of key value of ith element and key value of nth element in multi-dimensional feature vector1、x2、xnRespectively a first element, a second element and an nth element of the multidimensional characteristic vector matrix.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a crime space-time prediction method and a system, wherein a map of a region to be predicted is subjected to grid division, crime case data are processed and predicted by taking each grid as a unit, a multi-granularity convolution neural network data processing model is utilized to obtain crime case number sequences, weather sequences, date sequences and region sequences of each grid in a preset historical period, then a GloVe natural language processing method is utilized to express each sequence as a vector matrix, all the vector matrices are spliced into a multi-dimensional feature vector matrix of each grid according to dates, a weighted multi-dimensional feature vector matrix of each grid is determined based on a dynamic fusion algorithm model of a self-attention machine system, a multi-window encoder is utilized to encode information of the weighted multi-dimensional feature vector matrix, encoding features captured by windows with different lengths are obtained to serve as a training set training classifier, and then a trained classifier is obtained, and finally, predicting the number of crime cases of each grid on a target date by using the trained classifier, thereby realizing efficient development of space-time analysis and prediction on crimes.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart illustrating a crime spatiotemporal prediction method according to the present invention;
fig. 2 is a schematic diagram of a multi-window encoder provided by the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for predicting a crime space-time to efficiently carry out space-time analysis and prediction on a crime.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
The invention is based on the problem that the deep analysis, research and judgment and deep application of data in the prior art are very weak, establishes an effective mathematical model when introducing artificial intelligence technology, particularly deep neural network technology, into financial crimes, accurately and efficiently carries out space-time analysis and prediction on financial crimes in the district, changes the traditional crime analysis passive coping mode, and enables the actual combat department to predict in advance, deploy in advance and prevent in advance.
The invention provides a crime space-time prediction method, as shown in figure 1, the prediction method comprises the following steps:
the method comprises the following steps of firstly, carrying out structuralization processing on unstructured data in an information platform by using a geographic information technology, realizing preprocessing of financial invasion crime data, cleaning the data by using a core task of the preprocessing, removing noise data, and extracting effective data, wherein the core task of the preprocessing is as follows:
s101, carrying out grid division on a map of an area to be predicted, and acquiring the time of criminal case occurrence of each grid in a preset historical period;
after the map is subjected to grid division, the longitude and latitude range of each grid can be obtained, and the grid to which the criminal case belongs can be determined according to the longitude and latitude information of the criminal case occurrence place.
The map of the area to be predicted is subjected to grid division, and each grid is taken as a unit to respectively process and predict crime case data, so that the comprehensive monitoring and crime case occurrence prevention of the area to be predicted are facilitated.
S102, determining the number of crime cases of each grid in a preset historical period according to the occurrence time of the crime cases;
after the step S102, the data outlier is removed by using an absolute median difference method (a method for removing outliers commonly used in statistics), and the specific steps are as follows:
determining median X of all crime cases in each grid in preset historical periodmedium(ii) a When the number of all crime cases is even, averaging and rounding the two middle data to be used as a median;
calculating absolute deviation values of the number of the criminal cases and the median of each grid in a preset historical period;
determining median MAD of all absolute deviation values;
median X according to crime case numbermediumAnd a median MAD of absolute deviation value, determining the range of the number of the crime cases as [ Xmedium-n×MAD,Xmedium+n×MAD]；
The number of crime cases is not in [ X ]medium-n×MAD,Xmedium+n×MAD]The value of the number of crime cases of the grids in the range is set to be 0;
where n is a constant scaling factor, preferably, n is 1.4826. The value of n depends on the type of distribution of the data set, with the aim of setting a confidence range, eliminating values that are not within this range.
Secondly, a natural language processing method is used to construct a training set, and the process is as follows:
s103, respectively filling the number of crime cases, the weather and the date of each day of each grid in a preset historical period and the region to which each grid belongs into sequences with the same length of days in a day unit by using a multi-granularity convolutional neural network data processing model to obtain the number sequence, the weather sequence, the date sequence and the region sequence of the crime cases of each grid in the preset historical period;
in prior art methods, different training data are required to be constructed to train a plurality of different models using different time windows for prediction. For example, assuming that seven-day historical data is used for prediction, training data padding (padding) needs to be unified into a seven-day format, such as {1: [2,0,3,1,7,6,3],2: [1,2,1,0,0,3,1 }, wherein 1 and 2 represent area labels, and the list represents the historical records of crime case number in the area. In the traditional method, historical crime data needs to be sorted into a sequence with the length of 7 to train a corresponding model to realize prediction by adopting seven-day historical data. If eight days of historical data are needed for prediction, the training data needs to be rearranged to unify training data padding (padding) into a sequence of length 8 to retrain the model.
In the invention, in order to solve the problems, an innovative multi-granularity convolutional neural network data processing model (OneForALL data processing model) is provided for the construction of the training data set. The special label "UNK" is used to indicate missing historical data. For example, if we want to make a prediction using 5 to 15 days of historical data, all training data padding (padding) is unified into a length-15 sequence (the length-15 sequence contains all historical data within 15 days) when generating training data, and historical data for consecutive 5 to 10 days is randomly removed from the left side of the list and replaced with a special label "UNK". Thus, 5 to 15 days of training data are available, and the model obtained by training can be predicted by using 5 to 15 days of historical data. For example, given 15 days of historical data {1: [2,0,3,1,7,6,3,5,2,0,3,1,7,6,3],2: [1,2,1,0,0,3,1,4,2,0,3,1,7,6,3 }, the data for 5 to 10 consecutive days in this example would be randomly removed and replaced with a UNK. {1: [ UNK, UNK, UNK, UNK, UNK, UNK, UNK, UNK, 2,0,3,1,7,6,3],2: [ UNK, UNK, UNK, UNK, UNK, 1,2,1,0,0,3,1 }. Because the model training data generated by the method contains historical data in different time ranges, the trained model has the capability of predicting according to the data in different historical ranges. Under test, for example, given {1: [2,0,3,1,7,6,3],2: [1,2,1,0,0,3,1] }, the historical data in each region is filled (padding) into a length 15 list {1: [ UNK, …,2,0,3,1,7,6,3],2: [ UNK, …,1,2,1,0,0,3,1] }, and the resulting data is processed as follows:
case number: [ UNK, UNK, UNK, UNK, UNK, UNK, UNK, UNK, 2,0,3,1,7,6,3]
Weather: [ UNK, UNK, UNK, UNK, UNK, UNK, UNK, UNK, sunny day, cloudy and cloudy, light rain, sunny day ]
Date: [ UNK, UNK, UNK, UNK, UNK, UNK, UNK, UNK, 6.11,6.12,6.13,6.14,6.15,6.16,6.17]
Region: [ UNK, UNK, UNK, UNK, UNK, UNK, UNK, UNK, Haishes ].
The specific step of S103 is summarized as follows:
expressing the number of crime cases per day in a preset historical period as an initial sequence of the number of cases according to dates;
representing the weather of each day in a preset historical period as an initial weather sequence according to the date;
representing the dates of each day in a preset historical period as an initial date sequence according to the dates;
representing the region to which each grid belongs in a preset historical period as a region initial sequence in a day unit according to the date;
and filling the case number initial sequence, the weather initial sequence, the date initial sequence and the region initial sequence into a sequence with a preset number of days by using a multi-granularity convolution neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in a preset historical period.
S104, respectively representing the number sequence, the weather sequence, the date sequence and the region sequence of the criminal cases as vector matrixes with the same dimensionality by using a GloVe natural language processing method, and splicing all the vector matrixes into a multi-dimensional characteristic vector matrix of each grid according to dates;
in the step, a word vector representation (GloVeembedding) model based on a GloVeembedding method in a natural language processing method is adopted to process words in a sequence, and the GloVeembedding model is trained: and constructing a characteristic word list for weather, date and region. Taking weather as an example, the vocabulary includes "light rain", "medium rain", "fine", etc., and these vocabularies are put into the weather feature vocabulary and are randomly initialized to vectors of fixed dimensions. Taking weather as an example, vectors corresponding to all weather are initialized randomly for all possible values of the weather, such as sunny days, cloudy days, and the like, and for example, 5 dimensions are set as dimensions of the weather vectors. All weather can thus be represented in the form of 5-dimensional vectors, which are initially initialized randomly, and for example [ UNK, sunny day, cloudy, rainy, sunny day ] this sequence can be represented as a 15 x 5-dimensional matrix: [ [0.01,0.003,0.015,0.223, -0.732], … [0.123,0.325,0.522,0.532,0.579] ], the model is trained by using a weather feature vocabulary and all weather corresponding initialization vectors, and parameters of the vectors are continuously updated in the training process.
After the GloVeembedding model was trained:
respectively converting the number of crime cases in a crime case number sequence, the weather in a weather sequence, the date in a date sequence and the region in a region sequence into vectors with the same dimensionality by using a trained GloVeembedding model;
all vectors in the crime case number sequence form a crime case number vector matrix, all vectors in the weather sequence form a weather vector matrix, all vectors in the date sequence form a date vector matrix, and all vectors in the region sequence form a region vector matrix;
and splicing the number vector matrix, the weather vector matrix, the number vector of the same date in the date vector matrix and the region vector matrix, the weather vector, the date vector matrix and the region vector to form the multi-dimensional characteristic vector matrix of each grid.
S105, determining a weighted multi-dimensional feature vector matrix of each grid by utilizing a dynamic fusion algorithm model based on a self-attention mechanism according to the multi-dimensional feature vector matrix of each grid;
the CNN-based feature encoder can only capture feature dependency in a window range, while the LSTM-based feature encoder can capture feature information in a larger range, but the encoder still has the problem of insufficient long-distance dependency capture capability. In order to solve the problem, the invention provides a dynamic fusion algorithm model based on a self-attentive mechanism (self-attentive mechanism), and obtains the dependency relationship in different time intervals by fusing the criminal history information of days in a period of time.
Taking a vector corresponding to each date in the multi-dimensional feature vector matrix as an element of the multi-dimensional feature vector matrix, and assigning each element with a key value;
respectively calculating the similarity between the key value of the ith element and the key value of each element in the multi-dimensional feature vector;
normalizing all the similarity by utilizing softmax to obtain a normalized value of the similarity;
according to a formula x'i＝k1x1+k2x2+…+knxnObtaining the weighting vector x of the ith elementi′；
All the weighted vectors form a weighted multidimensional characteristic vector matrix of each grid;
wherein k is1Normalized value, k, of the similarity of the key value of the ith element to the key value of the 1 st element in the multi-dimensional feature vector2Normalized value, k, of the similarity of the key value of the ith element to the key value of the 2 nd element in the multi-dimensional feature vectornNormalized value, x, of similarity of key value of ith element and key value of nth element in multi-dimensional feature vector1、x2、xnRespectively a first element, a second element and an nth element of the multidimensional characteristic vector matrix.
S106, a multi-window encoder is adopted to carry out information encoding on the weighted multi-dimensional characteristic vector matrix, and encoding characteristics captured by windows with different lengths are obtained;
after multi-view characteristic dynamic fusion is carried out, the dependency relationship among historical days is obtained, and in order to capture higher-level characteristics, a multi-window encoder is adopted for information encoding, namely, windows with different sizes of 2, 3 and 4 are adopted to capture the characteristics of 2-gram, 3-gram and 4-gram. In conventional convolutional neural networks, the user extracts features through a fixed window size convolution kernel. The fixed window size convolution kernels would be in the sequence x in left-to-right order1,x2,x3,......x15]Sliding, e.g., when the window size is 3, it extracts features that can only be considered for 3 days. Correspondingly, when the window size is 2, the considered information can only be information within 2 days, and the patent fuses a plurality of windows with different sizes to extract information with different granularities, and finally obtains the coding characteristics of the multi-window size, namely obtains more various dependent information at the same time, as shown in fig. 2.
Thirdly, training the model
S107, training the classifier by taking the coding features captured by the windows with different lengths as a training set to obtain a trained classifier; in the model training process, the invention adopts the stochastic gradient descent algorithm to continuously update and adjust the parameters of the model, so that the model prediction is more and more accurate. The specific process is as follows:
updating the parameters of the classifier by using a random gradient descent algorithm, and taking the parameter with the minimum loss function value as the optimal parameter of the classifier to obtain the trained classifier; the loss function is H (p, q) ═ Σ p (x) logq (x), where H (p, q) is the loss function, p (x) is the probability distribution of the number of real crime cases on the history date, and q (x) is the probability distribution of the number of predicted crime cases on the history date.
And applying a softmax function to convert the number of the issues of the target date into a probability distribution form.
From the formula of the loss function, it can be seen that a smaller value indicates a better model.
Updating parameters by a random gradient descent algorithm: calculating a loss function to calculate the partial derivative of the model parameter, determining the descending size and gradient direction of the parameter, and continuously adjusting the size of the model parameter along the gradient direction.
Fourthly, forecasting the occurrence quantity of crime cases
And S108, inputting the target date, the weather of the target date and the region to which each grid belongs into the trained classifier, and predicting the number of crime cases of each grid of the region to be predicted on the target date.
The invention provides a crime analysis and prediction model based on deep learning based on urgent actual combat requirements of a first-line criminal investigation department so as to be applied to actual combat. The method has the advantages that firstly, a set of OneFrall algorithm model based on the deep neural network is constructed, deep learning and crime space-time prediction are deeply fused, and a technical thought is provided for AI + crime prediction; and secondly, the model provides support for a first-line actual combat department to carry out crime prediction analysis, provides decision support for preventing crime attack and burglary and provides a corresponding prediction tool.
The invention can further accelerate the case handling efficiency and optimize the police resource allocation, thereby improving the social and public safety level.
The crime space-time prediction method provided by the invention is integrated to form a systematic analysis prediction platform, so that the visual presentation is formed while the crime data prediction is realized. Therefore, the present invention also provides a crime spatiotemporal prediction system, wherein the prediction system comprises: the system comprises a crime case occurrence time acquisition module, a crime case number determination module, a sequence acquisition module, a multi-dimensional feature vector matrix determination module, a weighted multi-dimensional feature vector matrix determination module, a coding feature capture module, a trained classifier acquisition module and a crime case occurrence number prediction module.
The criminal case occurrence time acquisition module is used for dividing grids of a map of an area to be predicted and acquiring criminal case occurrence time of each grid in a preset historical period;
the system comprises a crime case number determining module, a crime case number determining module and a crime case number determining module, wherein the crime case number determining module is used for determining the number of crime cases of each grid in a preset historical period according to the occurrence time of the crime cases;
the sequence obtaining module is used for respectively filling the number of crime cases, the weather and the date of each day of each grid in a preset historical period and the region to which each grid belongs into sequences with the same length of days by taking the day as a unit by utilizing a multi-granularity convolutional neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in the preset historical period;
the multi-dimensional characteristic vector matrix determining module is used for respectively representing the crime case number sequence, the weather sequence, the date sequence and the region sequence as vector matrixes with the same dimension by using a GloVe natural language processing method, and splicing all the vector matrixes into a multi-dimensional characteristic vector matrix of each grid according to the dates;
the weighted multi-dimensional eigenvector matrix determining module is used for determining the weighted multi-dimensional eigenvector matrix of each grid by utilizing a dynamic fusion algorithm model based on a self-attention mechanism according to the multi-dimensional eigenvector matrix of each grid;
the coding feature capturing module is used for carrying out information coding on the weighted multidimensional feature vector matrix by adopting a multi-window coder to obtain coding features captured by windows with different lengths;
the trained classifier obtaining module is used for training the classifier by taking the code characteristics captured by the windows with different lengths as a training set to obtain the trained classifier;
and the crime case occurrence number prediction module is used for inputting the target date, the weather of the target date and the region to which each grid belongs into the trained classifier and predicting the crime case occurrence number of each grid of the region to be predicted on the target date.
The sequence obtaining module specifically comprises: an initial sequence submodule of case number, an initial sequence submodule of weather, an initial sequence submodule of date, an initial sequence submodule of region and a sequence determining submodule.
The case number initial sequence submodule is used for representing the number of crime cases per day in a preset historical period as a case number initial sequence according to dates;
the weather initial sequence submodule is used for representing the weather of each day in a preset historical period as a weather initial sequence according to the date;
the date initial sequence submodule is used for expressing the dates of each day in the preset historical period as a date initial sequence according to the dates;
the region initial sequence submodule is used for expressing the region to which each grid belongs in a preset historical period as a region initial sequence in a day unit according to the date;
and the sequence determining submodule is used for filling the case number initial sequence, the weather initial sequence, the date initial sequence and the region initial sequence into a sequence with the length of preset days by using a multi-granularity convolutional neural network data processing model, so as to obtain the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in a preset historical period.
The multi-dimensional feature vector matrix determining module specifically includes: the vector transformation submodule, the vector matrix construction submodule and the multi-dimensional characteristic vector matrix construction submodule.
The vector conversion submodule is used for converting the number of crime cases in a crime case number sequence, the weather in a weather sequence, the date in a date sequence and the region in a region sequence into vectors with the same dimensionality respectively by utilizing a trained word vector representation model based on a GloVe method;
the vector matrix forming submodule is used for forming a crime case number vector matrix by all vectors in a crime case number sequence, forming a weather vector matrix by all vectors in a weather sequence, forming a date vector matrix by all vectors in a date sequence and forming a region vector matrix by all vectors in a region sequence;
and the multi-dimensional eigenvector matrix forming submodule is used for splicing the number vector matrix of the crime case, the weather vector matrix, the number vector of the date vector matrix and the region vector matrix on the same date, the weather vector, the date vector matrix and the region vector to form the multi-dimensional eigenvector matrix of each grid.
The weighted multidimensional feature vector matrix determining module specifically comprises: the key value assignment submodule, the similarity meter operator module, the similarity normalization value acquisition submodule, the weighting vector acquisition submodule and the weighting multidimensional characteristic vector matrix constitute submodules.
The key value assignment submodule is used for taking the vector corresponding to each date in the multi-dimensional characteristic vector matrix as an element of the multi-dimensional characteristic vector matrix and assigning each element with a key value;
the similarity calculation operator module is used for respectively calculating the similarity between the key value of the ith element and the key value of each element in the multi-dimensional feature vector;
the similarity normalization value acquisition submodule is used for normalizing all the similarities by utilizing softmax to obtain a similarity normalization value;
a weight vector obtaining submodule for obtaining a weight vector according to the formula xi′＝k1x1+k2x2+…+knxnObtaining the weighting vector x of the ith elementi′；
A weighted multidimensional characteristic vector matrix forming submodule for forming a weighted multidimensional characteristic vector matrix of each grid by all weighted vectors;
wherein k is1Normalized value, k, of the similarity of the key value of the ith element to the key value of the 1 st element in the multi-dimensional feature vector2Normalized value, k, of the similarity of the key value of the ith element to the key value of the 2 nd element in the multi-dimensional feature vectornNormalized value, x, of similarity of key value of ith element and key value of nth element in multi-dimensional feature vector1、x2、xnRespectively a first element, a second element and an nth element of the multidimensional characteristic vector matrix.
The system has the advantages of simplicity, convenience, practicability and easy operability, and aims to provide an intelligent analysis and judgment tool for carrying out space-time prediction for financial invasion crimes for actual combat departments, the quantity of financial invasion crime outbursts in a future period of time in a jurisdiction and a hotspot outburst area can be predicted through the platform tool, the actual combat departments can conveniently predict future outburst conditions in the jurisdiction in advance, prevention and control measures can be carried out in advance, and crime prevention and fighting efficiency can be improved.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (10)
1. A crime spatiotemporal prediction method, the prediction method comprising:
carrying out grid division on a map of an area to be predicted, and acquiring the time of criminal case occurrence of each grid in a preset historical period;
determining the number of crime cases of each grid in a preset historical period according to the occurrence time of the crime cases;
respectively filling the number of crime cases, the weather and the date of each day of each grid in a preset historical period and the region to which each grid belongs into sequences with the same length of days by taking the day as a unit by utilizing a multi-granularity convolutional neural network data processing model to obtain the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in the preset historical period;
respectively representing the crime case number sequence, the weather sequence, the date sequence and the region sequence as vector matrixes with the same dimensionality by using a GloVe natural language processing method, and splicing all the vector matrixes into a multi-dimensional characteristic vector matrix of each grid according to dates;
determining a weighted multidimensional characteristic vector matrix of each grid by utilizing a dynamic fusion algorithm model based on a self-attention mechanism according to the multidimensional characteristic vector matrix of each grid;
adopting a multi-window encoder to perform information encoding on the weighted multi-dimensional eigenvector matrix to obtain encoding characteristics captured by windows with different lengths;
training a classifier by taking the code characteristics captured by the windows with different lengths as a training set to obtain a trained classifier;
inputting the target date, the weather of the target date and the region to which each grid belongs into the trained classifier, and predicting the number of crime cases of each grid of the region to be predicted on the target date.
2. The crime spatiotemporal prediction method according to claim 1, wherein the determining the number of crime cases per day in each grid in a preset historical period according to the time of occurrence of the crime cases further comprises:
determining median X of all crime cases in each grid in preset historical periodmedium；
Calculating the absolute deviation value between the number of the criminal cases of each grid in a preset historical period and the median;
determining median MAD of all absolute deviation values;
according to median X of the number of the crime casesmediumAnd the median MAD of the absolute deviation value, and determining the range of the number of the crime cases as [ Xmedium-n×MAD,Xmedium+n×MAD]；
The number of crime cases is not in [ X ]medium-n×MAD,Xmedium+n×MAD]The value of the number of crime cases of the grids in the range is set to be 0;
where n is a constant scaling factor.
3. The crime spatiotemporal prediction method according to claim 1, wherein the obtaining of the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in the preset history period by filling the crime case number, the weather and the date of each day of each grid in the preset history period and the region to which each grid belongs in a sequence of the same length of days in a unit of day by using a multi-granularity convolutional neural network data processing model specifically comprises:
expressing the number of crime cases per day in a preset historical period as an initial sequence of the number of cases according to dates;
representing the weather of each day in a preset historical period as an initial weather sequence according to the date;
representing the dates of each day in a preset historical period as an initial date sequence according to the dates;
representing the region to which each grid belongs in a preset historical period as a region initial sequence in a day unit according to the date;
and filling the case number initial sequence, the weather initial sequence, the date initial sequence and the region initial sequence into a sequence with a preset number of days by using a multi-granularity convolutional neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in a preset historical period.
4. The crime spatiotemporal prediction method according to claim 1, wherein the crime case number sequence, the weather sequence, the date sequence and the region sequence are respectively represented as vector matrices of the same dimension by using a GloVe natural language processing method, and all vector matrices are spliced into a multidimensional feature vector matrix of each grid according to dates, specifically comprising:
converting the number of crime cases in the crime case number sequence, the weather in the weather sequence, the date in the date sequence and the region in the region sequence into vectors with the same dimensionality respectively by using a trained word vector representation model based on a GloVe method;
all vectors in the crime case number sequence form a crime case number vector matrix, all vectors in the weather sequence form a weather vector matrix, all vectors in the date sequence form a date vector matrix, and all vectors in the region sequence form a region vector matrix;
and splicing the number vector matrix, the weather vector matrix, the number vector of the same date in the date vector matrix and the region vector matrix, the weather vector, the date vector matrix and the region vector to form the multi-dimensional characteristic vector matrix of each grid.
5. The method according to claim 1, wherein the determining the weighted multidimensional eigenvector matrix for each grid by using a dynamic fusion algorithm model based on a self-attention mechanism according to the multidimensional eigenvector matrix for each grid specifically comprises:
taking a vector corresponding to each date in the multi-dimensional feature vector matrix as an element of the multi-dimensional feature vector matrix, and assigning each element with a key value;
respectively calculating the similarity between the key value of the ith element and the key value of each element in the multi-dimensional feature vector;
normalizing all the similarity by utilizing softmax to obtain a normalized value of the similarity;
according to the formula xi′＝k1x1+k2x2+…+knxnObtaining the weighting vector x of the ith elementi′；
All the weighted vectors form a weighted multidimensional characteristic vector matrix of each grid;
wherein k is1Normalized value, k, of the similarity of the key value of the ith element to the key value of the 1 st element in the multi-dimensional feature vector2Normalized value, k, of the similarity of the key value of the ith element to the key value of the 2 nd element in the multi-dimensional feature vectornNormalized value, x, of similarity of key value of ith element and key value of nth element in multi-dimensional feature vector1、x2、xnRespectively a first element, a second element and an nth element of the multidimensional characteristic vector matrix.
6. The method according to claim 1, wherein the training a classifier using the coding features of the multi-window size as a training set to obtain a trained classifier specifically comprises:
updating the parameters of the classifier by using a random gradient descent algorithm, and taking the parameter with the minimum loss function value as the optimal parameter of the classifier to obtain the trained classifier; the loss function is H (p, q) ═ Σ p (x) logq (x), where H (p, q) is the loss function, p (x) is the probability distribution of the number of real crime cases on the history date, and q (x) is the probability distribution of the number of predicted crime cases on the history date.
7. A criminal spatiotemporal prediction system, the prediction system comprising:
the criminal case occurrence time acquisition module is used for dividing grids of a map of an area to be predicted and acquiring criminal case occurrence time of each grid in a preset historical period;
the number determining module of the crime cases is used for determining the number of the crime cases of each grid in a preset historical period according to the occurrence time of the crime cases;
the sequence obtaining module is used for respectively filling the number of crime cases, the weather and the date of each day of each grid in a preset historical period and the region to which each grid belongs into sequences with the same length of days by taking the day as a unit by utilizing a multi-granularity convolutional neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in the preset historical period;
the multi-dimensional characteristic vector matrix determining module is used for respectively representing the crime case number sequence, the weather sequence, the date sequence and the region sequence as vector matrixes with the same dimension by using a GloVe natural language processing method, and splicing all the vector matrixes into a multi-dimensional characteristic vector matrix of each grid according to the dates;
the weighted multi-dimensional eigenvector matrix determining module is used for determining the weighted multi-dimensional eigenvector matrix of each grid by utilizing a dynamic fusion algorithm model based on a self-attention mechanism according to the multi-dimensional eigenvector matrix of each grid;
the coding feature capturing module is used for carrying out information coding on the weighted multidimensional feature vector matrix by adopting a multi-window coder to obtain coding features captured by windows with different lengths;
the trained classifier obtaining module is used for taking the code characteristics captured by the windows with different lengths as a training set to train the classifier so as to obtain a trained classifier;
and the crime case occurrence number prediction module is used for inputting the target date, the weather of the target date and the region to which each grid belongs into the trained classifier and predicting the crime case occurrence number of each grid of the region to be predicted on the target date.
8. The system of claim 7, wherein the sequence obtaining module specifically comprises:
the case number initial sequence submodule is used for representing the number of crime cases per day in a preset historical period as a case number initial sequence according to dates;
the weather initial sequence submodule is used for representing the weather of each day in a preset historical period as a weather initial sequence according to the date;
the date initial sequence submodule is used for expressing the dates of each day in the preset historical period as a date initial sequence according to the dates;
the region initial sequence submodule is used for expressing the region to which each grid belongs in a preset historical period as a region initial sequence in a day unit according to the date;
and the sequence determining submodule is used for filling the case number initial sequence, the weather initial sequence, the date initial sequence and the region initial sequence into a sequence with a preset number of days by using a multi-granularity convolutional neural network data processing model, and obtaining the crime case number sequence, the weather sequence, the date sequence and the region sequence of each grid in a preset historical period.
9. The system according to claim 7, wherein the multi-dimensional eigenvector matrix determination module specifically comprises:
the vector conversion submodule is used for converting the number of crime cases in the crime case number sequence, the weather in the weather sequence, the date in the date sequence and the region in the region sequence into vectors with the same dimension respectively by using a trained word vector representation model based on a GloVe method;
the vector matrix forming submodule is used for forming a crime case number vector matrix by all vectors in the crime case number sequence, forming a weather vector matrix by all vectors in the weather sequence, forming a date vector matrix by all vectors in the date sequence and forming a region vector matrix by all vectors in the region sequence;
and the multi-dimensional eigenvector matrix forming submodule is used for splicing the number vector matrix of the crime case, the weather vector matrix, the number vector of the date vector matrix and the region vector matrix on the same date, the weather vector, the date vector matrix and the region vector to form the multi-dimensional eigenvector matrix of each grid.
10. The crime spatiotemporal prediction method according to claim 1, characterized in that the weighted multidimensional eigenvector matrix determination module specifically comprises:
the key value assignment submodule is used for taking a vector corresponding to each date in the multi-dimensional characteristic vector matrix as an element of the multi-dimensional characteristic vector matrix and assigning each element with a key value;
the similarity calculation operator module is used for respectively calculating the similarity between the key value of the ith element and the key value of each element in the multi-dimensional feature vector;
the similarity normalization value acquisition submodule is used for normalizing all the similarities by utilizing softmax to obtain a similarity normalization value;
obtaining submodule of weighted vector according to formula x'i＝k1x1+k2x2+…+knxnObtaining a weighting vector x 'of the ith element'i；
A weighted multidimensional characteristic vector matrix forming submodule for forming a weighted multidimensional characteristic vector matrix of each grid by all weighted vectors;
wherein k is1Normalized value, k, of the similarity of the key value of the ith element to the key value of the 1 st element in the multi-dimensional feature vector2Normalized value, k, of the similarity of the key value of the ith element to the key value of the 2 nd element in the multi-dimensional feature vectornNormalized value, x, of similarity of key value of ith element and key value of nth element in multi-dimensional feature vector1、x2、xnRespectively a first element, a second element and an nth element of the multidimensional characteristic vector matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010828611.0A CN111898836A (en) | 2020-08-18 | 2020-08-18 | Crime space-time prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010828611.0A CN111898836A (en) | 2020-08-18 | 2020-08-18 | Crime space-time prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111898836A true CN111898836A (en) | 2020-11-06 |
Family
ID=73230154
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010828611.0A Withdrawn CN111898836A (en) | 2020-08-18 | 2020-08-18 | Crime space-time prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111898836A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112800158A (en) * | 2021-01-19 | 2021-05-14 | 吉林大学 | Vectorization representation method of geological map |
| CN113361855A (en) * | 2021-05-07 | 2021-09-07 | 浙江警官职业学院 | Short, medium and long-term risk warning method and device |
| CN113516556A (en) * | 2021-05-13 | 2021-10-19 | 支付宝(杭州)信息技术有限公司 | Method and system for predicting or training model based on multi-dimensional time series data |
| CN114581252A (en) * | 2022-03-03 | 2022-06-03 | 平安科技（深圳）有限公司 | Target case prediction method and device, electronic device and storage medium |
-
2020
- 2020-08-18 CN CN202010828611.0A patent/CN111898836A/en not_active Withdrawn
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112800158A (en) * | 2021-01-19 | 2021-05-14 | 吉林大学 | Vectorization representation method of geological map |
| CN113361855A (en) * | 2021-05-07 | 2021-09-07 | 浙江警官职业学院 | Short, medium and long-term risk warning method and device |
| CN113516556A (en) * | 2021-05-13 | 2021-10-19 | 支付宝(杭州)信息技术有限公司 | Method and system for predicting or training model based on multi-dimensional time series data |
| CN114581252A (en) * | 2022-03-03 | 2022-06-03 | 平安科技（深圳）有限公司 | Target case prediction method and device, electronic device and storage medium |
| CN114581252B (en) * | 2022-03-03 | 2024-04-05 | 平安科技（深圳）有限公司 | Target case prediction method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111898836A (en) | Crime space-time prediction method and system | |
| CN110647900B (en) | Intelligent safety situation prediction method, device and system based on deep neural network | |
| CN110995475B (en) | Power communication network fault detection method based on transfer learning | |
| CN109658695B (en) | Multi-factor short-term traffic flow prediction method | |
| CN111223301B (en) | Traffic flow prediction method based on graph attention convolution network | |
| CN109583565A (en) | Forecasting Flood method based on the long memory network in short-term of attention model | |
| CN110046787A (en) | A kind of urban area charging demand for electric vehicles spatio-temporal prediction method | |
| JP2013074695A (en) | Device, method and program for predicting photovoltaic generation | |
| CN109117894B (en) | Large-scale remote sensing image building classification method based on full convolution neural network | |
| CN110738355A (en) | urban waterlogging prediction method based on neural network | |
| CN115223049B (en) | Knowledge distillation and quantification method for large model compression of electric power scene edge calculation | |
| CN115587454A (en) | Traffic flow long-term prediction method and system based on improved Transformer model | |
| CN112910711A (en) | Wireless service flow prediction method, device and medium based on self-attention convolutional network | |
| CN110570035A (en) | people flow prediction system for simultaneously modeling space-time dependency and daily flow dependency | |
| CN114091361B (en) | Weather event-based transform model construction method | |
| CN116310350B (en) | Urban scene semantic segmentation method based on graph convolution and semi-supervised learning network | |
| CN111882157A (en) | Demand prediction method and system based on deep space-time neural network and computer readable storage medium | |
| CN113345236A (en) | Time-space traffic state prediction method based on Transformer network | |
| CN109447331A (en) | Mountain fire Risk Forecast Method based on stacking algorithm | |
| CN112016590A (en) | Prediction method combining sequence local feature extraction and depth convolution prediction model | |
| CN109829583B (en) | Mountain fire risk prediction method based on probability programming technology | |
| CN113192325B (en) | Traffic incident correlation analysis method and system | |
| CN117636183A (en) | Small sample remote sensing image classification method based on self-supervision pre-training | |
| CN115713044B (en) | Method and device for analyzing residual life of electromechanical equipment under multi-condition switching | |
| CN117115706A (en) | Video scene graph generation method based on multi-scale space-time attention network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20201106 |