A risk assessment and early warning method and device based on multi-dimensional information fusion

Abstract

The invention discloses a risk assessment and early warning method and device based on multi-dimensional information fusion, which belongs to the technical field of data processing. The invention adopts a data processing method based on T-SNE dimension reduction and K-Means clustering algorithm for multi-dimensional heterogeneous data of a scheduled place, thereby reducing the difficulty of data fusion of different dimensions and the problem that the nonlinear correlation of data of each dimension affects the prediction result; and adopts a method combining model learning with expert experience to conduct risk assessment and early warning, and uses K-means clustering algorithm and LSTM to conduct model assessment and prediction, which can fully explore the mathematical laws in the data, and combines expert experience for scoring, which increases the rationality and interpretability of the assessment and prediction results; adopts a confidence calculation and prediction method, and after the data processing module completes the risk prediction, only a high, medium or low risk value result is obtained, thereby providing a guarantee for making a reasonable and scientific decision in the scheduled place.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a risk assessment and early warning method and device based on multi-dimensional information fusion. Background Art

[0002] With the continuous optimization of machine learning algorithms, the application of artificial intelligence technology is becoming more and more extensive, penetrating into all aspects of finance, engineering and the judicial system. In recent years, artificial intelligence technology has also made great progress in risk assessment and early warning, but it still has some shortcomings in the application of reservation venues: due to the particularity and limitations of reservation venue data, there are problems such as insufficient data system and single prediction model when conducting analysis, and the correlation between data and risk cannot be fully explored.

[0003] Risk assessment and early warning of a reservation site refers to the assessment of the current risk scenario and risk level in the reservation site, as well as the prediction and assessment of various possible situations in the risk scenario. Smart reservation sites are a new and important application in the field of artificial intelligence in the current era. Under the smart scenario mode of reservation sites, reservation sites integrate regulatory transformation information resources and social information resources. How to integrate modern technologies such as big data, the Internet of Things, and artificial intelligence with various businesses of reservation sites, take quantitative and qualitative assessments of regulatory safety risks, improve the prevention capabilities of reservation sites in terms of regulatory safety, and promote innovation in the governance system of reservation sites are technical problems that need to be solved urgently. Summary of the invention

[0004] The present invention provides a risk assessment and early warning method and device based on multi-dimensional information fusion, which integrates business data of multiple dimensions including personnel, equipment, etc. in a predetermined place and uses T-SNE dimensionality reduction and K-means clustering algorithms for data processing through a combination of model training and expert experience. The risks of the predetermined place are assessed and predicted through data prediction using a long short-term memory neural network, which enhances the interpretability and rationality of the prediction results and can provide reference and support for regulators in risk prevention and disposal decisions.

[0005] The specific technical solutions provided by the present invention are as follows:

[0006] On the one hand, the present invention provides a risk assessment and early warning method based on multi-dimensional information fusion, comprising:

[0007] Collect data of various dimensions of personnel and equipment in the scheduled place, wherein the data includes basic information data of preset personnel, scoring and evaluation data, danger level data of preset personnel, behavioral benchmark data of preset personnel, number of blacklist and whitelist calls in the scheduled place, laser perimeter alarm data, security system alarm data, and video surveillance data;

[0008] Perform fusion preprocessing on the collected data, wherein the fusion preprocessing includes data filling, data cleaning, T-SNE dimensionality reduction processing, and K-Means clustering processing;

[0009] A priori knowledge base for risk assessment of scheduled places is constructed in combination with expert experience, and the risk level of the scheduled places is evaluated and predicted based on the priori knowledge base and the data after fusion preprocessing, wherein the risk level of the scheduled places includes high risk, medium risk and low risk.

[0010] Optionally, the evaluation and prediction of the risk level of a predetermined place based on the prior knowledge base and the fusion preprocessed data is specifically as follows:

[0011] Combining the data collected at the current moment with the historical data of the previous N days in the prior knowledge base, the data of various topics in the future period are predicted through the long short-term memory artificial neural network;

[0012] Input the historical data of each dimension into the time series LSTM model to predict the data of different dimensions in the future, and obtain the possible data of the future in the same format as the original data;

[0013] Based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, the risk level at the current moment is evaluated and the future risk level of the scheduled place is predicted.

[0014] Optionally, the risk level at the current moment and the future risk level of the scheduled place are evaluated based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, specifically:

[0015] After obtaining the business data at the current moment, it is input into the prior knowledge base to obtain the risk results of the scheduled venue in the current scenario;

[0016] Afterwards, the current business data and risk assessment results are input into the database. At the same time, combined with the current input data and the historical data of the previous N days in the database, the business data of each dimension in the future period is predicted separately through the long short-term memory artificial neural network, and the predicted data in the future period is input into the prior knowledge base to obtain the risk results of the scheduled venues in the future scenarios.

[0017] Optionally, the fusion preprocessing of the collected data is specifically:

[0018] After collecting business data of all dimensions, fill and clean the data each time, and then store it in the database;

[0019] Perform T-SNE dimensionality reduction on the stored data, and reduce our nonlinear related data to 2 dimensions through the T-SNE dimensionality reduction algorithm;

[0020] The data after dimensionality reduction are clustered using the K-Means clustering algorithm to obtain clustered data divided into three clusters, where the three clusters correspond to high, medium and low risks of the scheduled places, respectively. The corresponding relationship is determined by the experts' scoring based on their prior knowledge.

[0021] Optionally, the data after the dimensionality reduction process is clustered by a K-Means clustering algorithm to obtain clustered data divided into three clusters:

[0022] (1) Let t = 0 and randomly select 3 sample points as the initial cluster centers

[0023] (2) For a fixed cluster center in, For class G i The center of the sample X′ is calculated ij The distance to the cluster center is calculated as Assign each sample to the class with the closest center to form the clustering result C (t) ;

[0024] (3) Clustering result C (t) , calculate the mean of the samples in each current class as the new class center

[0025] (4) If the iteration converges or meets the stopping condition, output C * =C (t) ,get Three classes, otherwise t=t+1, return to step (2) and execute in a loop.

[0026] On the other hand, the present invention also provides a risk assessment and early warning device based on multi-dimensional information fusion, comprising:

[0027] A collection unit is configured to collect data of various dimensions of personnel and equipment in a predetermined place, wherein the data includes basic information data of preset personnel, scoring and evaluation data, risk level data of preset personnel, behavioral benchmark data of preset personnel, number of blacklist and whitelist calls in the predetermined place, laser perimeter alarm data, security system alarm data, and video surveillance data;

[0028] A processing unit, configured to perform fusion preprocessing on the collected data, wherein the fusion preprocessing includes data filling, data cleaning, T-SNE dimensionality reduction processing, and K-Means clustering processing;

[0029] The evaluation and early warning unit is configured to construct a priori knowledge base for risk assessment of predetermined places in combination with expert experience, and to evaluate and predict the risk level of the predetermined place based on the priori knowledge base and the data after fusion preprocessing, wherein the risk level of the predetermined place includes high risk, medium risk and low risk.

[0030] Optionally, the evaluation and early warning unit is specifically configured as follows:

[0031] Combining the data collected at the current moment with the historical data of the previous N days in the prior knowledge base, the data of various topics in the future period are predicted through the long short-term memory artificial neural network;

[0032] Input the historical data of each dimension into the time series LSTM model to predict the data of different dimensions in the future, and obtain the possible data of the future in the same format as the original data;

[0033] Based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, the risk level at the current moment is evaluated and the future risk level of the scheduled place is predicted.

[0034] Optionally, the risk level at the current moment and the future risk level of the scheduled place are evaluated based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, specifically:

[0035] After obtaining the business data at the current moment, it is input into the prior knowledge base to obtain the risk results of the scheduled venue in the current scenario;

[0036] Afterwards, the current business data and risk assessment results are input into the database. At the same time, combined with the current input data and the historical data of the previous N days in the database, the business data of each dimension in the future period is predicted separately through the long short-term memory artificial neural network, and the predicted data in the future period is input into the prior knowledge base to obtain the risk results of the scheduled venues in the future scenarios.

[0037] Optionally, the processing unit is specifically configured as follows:

[0038] After collecting business data of all dimensions, fill and clean the data each time, and then store it in the database;

[0039] Perform T-SNE dimensionality reduction on the stored data, and reduce our nonlinear related data to 2 dimensions through the T-SNE dimensionality reduction algorithm;

[0040] The data after dimensionality reduction are clustered using the K-Means clustering algorithm to obtain clustered data divided into three clusters, where the three clusters correspond to high, medium and low risks of the scheduled places, respectively. The corresponding relationship is determined by the experts' scoring based on their prior knowledge.

[0041] Optionally, the data after the dimensionality reduction process is clustered by a K-Means clustering algorithm to obtain clustered data divided into three clusters:

[0042] (1) Let t = 0 and randomly select 3 sample points as the initial cluster centers

[0043] (2) For a fixed cluster center in, For class G i The center of the sample X′ is calculated ij The distance to the cluster center is calculated as Assign each sample to the class with the closest center to form the clustering result C (t) ;

[0044] (3) Clustering result C (t) , calculate the mean of the samples in each current class as the new class center

[0045] (4) If the iteration converges or meets the stopping condition, output C * =C (t) ,get Three classes, otherwise t=t+1, return to step (2) and execute in a loop.

[0046] The beneficial effects of the present invention are as follows:

[0047] The embodiment of the present invention provides a risk assessment and early warning method and device based on multi-dimensional information fusion. The method and device adopt a data processing method based on T-SNE dimensionality reduction and K-Means clustering algorithm for the multi-dimensional heterogeneous data of the scheduled place, which reduces the difficulty of data fusion of different dimensions and the problem that the nonlinear correlation of data of each dimension affects the prediction result; and adopts a method combining model learning and expert experience to evaluate and warn risks. The K-means clustering algorithm and LSTM are used to evaluate and predict the model, which can fully explore the mathematical laws in the data, and the scoring combined with expert experience increases the rationality and interpretability of the evaluation and prediction results; the prediction of future risk scenarios is the prediction of the original data, not the prediction of the risk scenarios obtained through T-SNE dimensionality reduction. The data after dimensioning or the risk value is predicted directly. Business data of different dimensions have their own unique regularities. The method of predicting the original data can fully explore the different rules of each dimension instead of obtaining a fused result. The prediction of data in multiple dimensions also increases the fault tolerance of the model, and while improving the accuracy, it also increases the visualization ability of the model prediction process. At the same time, the confidence calculation prediction method is adopted. After the data processing module completes the risk prediction, it only obtains a high, medium, or low risk value result. The distance between the predicted data and the cluster center point is calculated through the Euclidean distance and standardized method to obtain the confidence of the prediction result, which can more intuitively quantify the prediction result and provide guarantee for making reasonable and scientific decisions in the reserved place. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 It is a flow chart of a risk assessment and early warning method based on multi-dimensional information fusion provided by an embodiment of the present invention;

[0050] Figure 2 It is a structural block diagram of a risk assessment and early warning device based on multi-dimensional information fusion provided by an embodiment of the present invention;

[0051] Figure 3 It is a schematic diagram of a priori knowledge base for risk assessment of a predetermined place for constructing multi-dimensional data fusion provided by an embodiment of the present invention;

[0052] Figure 4 A structural block diagram of a long short-term memory network algorithm provided by an embodiment of the present invention;

[0053] Figure 5This is a structural block diagram of the risk warning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0055] The following will be combined with the attached Figure 1 ~Attached Figure 5 , a risk assessment and early warning method and device based on multi-dimensional information fusion in an embodiment of the present invention are described in detail.

[0056] refer to Figure 1 and Figure 3 As shown, a risk assessment and early warning method based on multi-dimensional information fusion provided by an embodiment of the present invention includes:

[0057] 100: Collect data of various dimensions of personnel and equipment in the predetermined place, wherein the data includes basic information data of preset personnel, scoring and evaluation data, risk level data of preset personnel, behavioral benchmark data of preset personnel, number of calls in the black and white lists in the predetermined place, laser perimeter alarm data, security system alarm data, and video surveillance data;

[0058] Collect data of various dimensions of personnel and equipment in the scheduled place. In order to ensure the accuracy of the clustering results, the business data of the scheduled place for 30 days is collected from the scheduled place big data platform. The data is collected according to the maximum collection volume, that is, once every five minutes. In this way, a total of 30 days of data are collected, 288 pieces of data per day, a total of 30×288=8640 pieces of data. Each piece of data contains 11 dimensions of business data, including the basic situation of the preset personnel, scoring evaluation, the danger level of the preset personnel, the behavioral benchmark data of the preset personnel, the number of black and white list calls in the scheduled place, the laser perimeter alarm data, the existing security system alarm data, and the existing video surveillance data. The data format is X ij =[x1,x2,x3,x4,x5,x6,x7,x8,x9,x 10 , x 11 ]. Among them, X ij Represents the jth (0<j≤288) data of the ith (0<i≤30) day. After data preprocessing such as data filling and cleaning, the data set is obtained. And store it into MySQL database.

[0059] MySQL is an open source relational database management system (RDBMS) that uses the most commonly used database management language - Structured Query Language (SQL) for database management.

[0060] 200: performing fusion preprocessing on the collected data, wherein the fusion preprocessing includes data filling, data cleaning, T-SNE dimension reduction processing, and K-Means clustering processing;

[0061] Specifically, each time after collecting business data of all dimensions, the data is filled and cleaned, and then stored in the database; the stored data is processed by T-SNE dimensionality reduction, and our nonlinear related data is reduced to 2 dimensions through the T-SNE dimensionality reduction algorithm; the data after dimensionality reduction is clustered through the K-Means clustering algorithm to obtain clustered data divided into three clusters, where the three clusters of cluster data correspond to high, medium and low risks of the scheduled places, respectively, and the corresponding relationship is determined by the experts' scoring based on their prior knowledge.

[0062] It should be noted that T-SNE, the full name of which is T-distributed Stochastic Neighbor Embedding, is a nonlinear dimensionality reduction algorithm that is very suitable for reducing high-dimensional data to 2 or 3 dimensions for visualization. K-Means, also known as the K-means clustering algorithm, is the most commonly used clustering algorithm. The input of the algorithm is a sample set (or point set). Through this algorithm, samples can be clustered and samples with similar characteristics are grouped into one category.

[0063] The data after dimensionality reduction is clustered by K-Means clustering algorithm, and the clustered data divided into three clusters are obtained as follows:

[0064] (1) Let t = 0 and randomly select 3 sample points as the initial cluster centers

[0065] (2) For a fixed cluster center in, For class G i The center of the sample X′ is calculated ij The distance to the cluster center is calculated as Assign each sample to the class with the closest center to form the clustering result C (t) ;

[0066] (3) Clustering result C (t) , calculate the mean of the samples in each current class as the new class center

[0067] (4) If the iteration converges or meets the stopping condition, output C * =C (t) ,get Three classes, otherwise t=t+1, return to step (2) and execute in a loop.

[0068] The fusion preprocessing of the collected data is divided into two parts. The first part is T-SNE dimensionality reduction. T-SNE is a nonlinear dimensionality reduction method. Nonlinear dimensionality reduction algorithms usually pay more attention to maintaining similarity, making the distance between similar points in low-dimensional space smaller. T-SNE dimensionality reduction converts the similarity between data points into probability. The similarity in the original space is represented by Gaussian joint probability, and the similarity in the embedded space is represented by "Student T distribution". After each data in the fused data set X is reduced to two-dimensional space through T-SNE, a new data set is obtained. where X′ ij =[x′1, x′2], x′1, x′2 are the data items after dimensionality reduction.

[0069] The second step of fusion preprocessing of the collected data is to cluster the reduced-dimensional data using the K-Means clustering algorithm. The input of the K-Means algorithm is a sample set (or point set), through which the samples can be clustered, and samples with similar characteristics are grouped into one category. For each point, calculate the center point that is closest to all center points, and then assign the point to the cluster represented by the center point. After one iteration, recalculate the center point for each cluster, and then re-find the center point closest to each point. Repeat this cycle until there is no change in the clusters between the previous and next iterations. The meaning of K is to divide the sample set into K categories. For example, K is 3, corresponding to high, medium, and low risk values ​​for the predetermined location.

[0070] After obtaining the clustering results of the business data, the original data X and the clustering results C * Fusion constitutes labeled data For X ij The corresponding clustering results. The scheduled venue supervisor will combine the original data X ij The clustering results obtained Make judgments and correspond to high, medium and low risks. That is, the prior knowledge base is constructed and stored in the database.

[0071] 300: Construct a priori knowledge base for risk assessment of a predetermined place in combination with expert experience, and evaluate and predict the risk level of the predetermined place based on the priori knowledge base and the data after fusion preprocessing, wherein the risk level of the predetermined place includes high risk, medium risk and low risk.

[0072] Specifically, the data collected at the current moment and the historical data of the previous N days in the prior knowledge base are combined, and the data of various topics in the future period are predicted through the long short-term memory artificial neural network; the historical data of each dimension is input into the time series LSTM model to predict the data of different dimensions in the future period, and the possible data in the future period in the same format as the original data is obtained; according to the current data after the prior knowledge base and fusion preprocessing, and the predicted data in the future period, the risk level at the current moment is evaluated and the future risk level of the scheduled place is predicted.

[0073] It should be noted that the full name of the LSTM model is Long Short-Term Memory Network (LSTM), which is a special RNN model. It is mainly used to solve the gradient disappearance and gradient explosion problems in the long sequence training process. It is a deep learning model for solving data containing time dimensions. The data processing process of the embodiment of the present invention uses the T-SNE dimensionality reduction algorithm and the K-Means clustering algorithm to process the heterogeneous data of the predetermined place. The prior knowledge base is a database with labels formed by adding expert experience to the historical data through the data processing module for subsequent risk assessment and prediction.

[0074] Furthermore, after obtaining the business data at the current moment, it is input into the prior knowledge base to obtain the risk result of the scheduled place in the current scenario; then the business data at the current moment and the risk assessment result are input into the database, and at the same time, combined with the input data at the current moment and the historical data of the previous N days in the database, the business data of each dimension in the future period is predicted separately through the long short-term memory artificial neural network, and the predicted data in the future period is input into the prior knowledge base to obtain the risk result of the scheduled place in the future scenario.

[0075] When assessing the risk of a reservation site scenario, the business data of 11 dimensions at the current moment is obtained from the reservation site big data platform and integrated to obtain X n =[x1,x2,x3,x4,x5,x6,x7,x8,x9,x 10 , x 11 ], where x1 to x 11 They represent the basic information of the preset personnel, scoring evaluation, the preset personnel danger level, the preset personnel behavior benchmark data, the number of blacklist and whitelist calls in the preset place, the laser perimeter alarm data, the existing security system alarm data, and the existing video surveillance data. n Input into the above data processing module, after dimension reduction and clustering, the clustering result is obtained The risk assessment result is the high, medium and low risk corresponding to the expert's prior knowledge. The risk assessment result is returned to the big data platform of the scheduled place and displayed.

[0076] refer to Figure 4 As shown, when early warning of future risks is carried out, the input data at the current moment and the historical data of the previous N days in the database are combined to predict the various data in the future through the long short-term memory artificial neural network. The various situations of the scheduled venue scene have a certain time regularity. By inputting the historical data of each dimension into the time series LSTM model, the data of different dimensions in the future are predicted to obtain the possible future data in the same format as the original data. All recursive neural networks have chain repeating modules of neural networks. LSTM has the same structure, but the repeated modules have different structures, such as Figure 5 As shown. Unlike the single neural network layer of RNN, the LSTM model has four network layers that interact in a very special way. The network algorithm structure diagram of LSTM is shown in Figure 5 The implementation steps are as follows:

[0077] (1) Forget gate: The first step of LSTM is to decide what information to discard from the cell state. This decision is made by an S-shaped network layer called the “forget gate layer”. It receives h t-1 and x t And the cell state C t-1 For each number in the output value is between 0 and 1. 1 means "completely accept this" and 0 means "completely ignore this". The calculation formula is: m =σ f (W f ·[h t-1 , x t ]+b f )

[0078] (2) Input gate: The next step is to determine what new information needs to be saved in the cell state. This is divided into two parts. In the first part, an S-shaped network layer called the "input gate layer" determines what information needs to be updated. In the second part, a tanh-shaped network layer creates a new candidate value vector C′ t , which can be used to add to the cell state. In the next step we combine the two parts above to produce an update to the state. The calculation formula is: t =σ i (W i [h t-1 , x t ]+b i ); C′ t =tanh(W c [h t-1 , x t ]+b c ).

[0079] (3) Cell state update: Now update the old cell state C t-1 Update to C t . Multiply the old state by f t , used to forget things that you decide to forget. Then add i t ⊙C′ t This is the new candidate value, scaled by the update value decided for each state. The calculation formula is: C t =f t ⊙C t-1 +i t ⊙C′ t .

[0080] (4) Output gate: Finally, the output value needs to be determined. The output depends on the cell state, but it will be a "filtered" version. First, the sigmoid network layer is run to determine which parts of the cell state can be output. Then, the cell state is input into tanh (adjust the value to between -1 and 1) and multiplied by the output value of the sigmoid network layer, so that the desired output score can be output. The calculation formula is: t =σ0(W0[h t-1 , x t ]+b0);h t =o t ⊙tanh(C t ).

[0081] The dataset X of the original business data in the database is passed through LSTM to obtain the predicted value y=[y1, y2, y3, y4, y5, y6, y7, y8, y9, y 10 ,y 11 ], where y1 to y 11 Represents X n =[x1,x2,x3,x4,x5,x6,x7,x8,x9,x 10 , x 11 ] The corresponding predicted value, after getting the prediction, y is still reduced by T-SNE to get y′=[y′1, y′2], and then K-Means clustering is performed to get the clustering result The high, medium and low risks corresponding to the expert's prior knowledge are the results of the risk warning. After obtaining the warning level, the confidence level needs to be solved. The main steps are:

[0082] (1) The distance from the cluster center. When constructing the prior knowledge base, we obtained Cluster centers of three types of results in Then the distance between the predicted value y′=[y′1, y′2] after T-SNE dimension reduction and the center point of each cluster is:

[0083] (2) Confidence. The confidence of the predicted value for a certain risk level can be calculated by comparing it with the limit value of the point of that type. The calculation method is as follows: First, find the distance from each type of point to the cluster center through the prior knowledge base. The distance from the farthest point to the cluster center Then the predicted value y belongs to The confidence of the class is:

[0084] The confidence value range is p i ≤1. When the confidence level is greater than 0, it is credible, indicating that the predicted value y belongs to Class risk, and p i The closer it is to 1, the higher the credibility is, and the closer it is to 0, the lower the credibility is. When the confidence is less than 0, it is unreliable, indicating that the predicted value y does not belong to Class risk, and p i The smaller the The smaller the probability of such risk.

[0085] The embodiment of the present invention provides a risk assessment and early warning method and device based on multi-dimensional information fusion. The method and device adopt a data processing method based on T-SNE dimensionality reduction and K-Means clustering algorithm for the multi-dimensional heterogeneous data of the scheduled place, which reduces the difficulty of data fusion of different dimensions and the problem that the nonlinear correlation of data of each dimension affects the prediction result; and adopts a method combining model learning and expert experience to evaluate and warn risks. The K-means clustering algorithm and LSTM are used to evaluate and predict the model, which can fully explore the mathematical laws in the data, and the scoring combined with expert experience increases the rationality and interpretability of the evaluation and prediction results; the prediction of future risk scenarios is the prediction of the original data, not the prediction of the risk scenarios obtained through T-SNE dimensionality reduction. The data after dimensioning or the risk value is predicted directly. Business data of different dimensions have their own unique regularities. The method of predicting the original data can fully explore the different rules of each dimension instead of obtaining a fused result. The prediction of data in multiple dimensions also increases the fault tolerance of the model, and while improving the accuracy, it also increases the visualization ability of the model prediction process. At the same time, the confidence calculation prediction method is adopted. After the data processing module completes the risk prediction, it only obtains a high, medium, or low risk value result. The distance between the predicted data and the cluster center point is calculated through the Euclidean distance and standardized method to obtain the confidence of the prediction result, which can more intuitively quantify the prediction result and provide guarantee for making reasonable and scientific decisions in the reserved place.

[0086] Based on the same inventive concept, Figure 2 As shown, the embodiment of the present invention also provides a risk assessment and early warning device based on multi-dimensional information fusion, including:

[0087] The collection unit 110 is configured to collect data of various dimensions of personnel and equipment in the predetermined place, wherein the data includes basic information data of preset personnel, scoring and evaluation data, risk level data of preset personnel, behavioral benchmark data of preset personnel, number of blacklist and whitelist calls in the predetermined place, laser perimeter alarm data, security system alarm data, and video surveillance data;

[0088] The processing unit 120 is configured to perform fusion preprocessing on the collected data, wherein the fusion preprocessing includes data filling, data cleaning, T-SNE dimensionality reduction processing, and K-Means clustering processing;

[0089] The evaluation and early warning unit 130 is configured to construct a priori knowledge base for risk assessment of predetermined places in combination with expert experience, and to evaluate and predict the risk level of the predetermined place based on the priori knowledge base and the data after fusion preprocessing, wherein the risk level of the predetermined place includes high risk, medium risk and low risk.

[0090] Optionally, the evaluation and early warning unit 130 is specifically configured as follows:

[0091] Combining the data collected at the current moment with the historical data of the previous N days in the prior knowledge base, the data of various topics in the future period are predicted through the long short-term memory artificial neural network;

[0092] Input the historical data of each dimension into the time series LSTM model to predict the data of different dimensions in the future, and obtain the possible data of the future in the same format as the original data;

[0093] Based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, the risk level at the current moment is evaluated and the future risk level of the scheduled place is predicted.

[0094] Optionally, the risk level at the current moment and the future risk level of the scheduled place are evaluated based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, specifically:

[0095] After obtaining the business data at the current moment, it is input into the prior knowledge base to obtain the risk results of the scheduled venue in the current scenario;

[0096] Afterwards, the current business data and risk assessment results are input into the database. At the same time, combined with the current input data and the historical data of the previous N days in the database, the business data of each dimension in the future period is predicted separately through the long short-term memory artificial neural network, and the predicted data in the future period is input into the prior knowledge base to obtain the risk results of the scheduled venues in the future scenarios.

[0097] Optionally, the processing unit 120 is specifically configured as follows:

[0098] After collecting business data of all dimensions, fill and clean the data each time, and then store it in the database;

[0099] Perform T-SNE dimensionality reduction on the stored data, and reduce our nonlinear related data to 2 dimensions through the T-SNE dimensionality reduction algorithm;

[0100] The data after dimensionality reduction are clustered using the K-Means clustering algorithm to obtain clustered data divided into three clusters, where the three clusters correspond to high, medium and low risks of the scheduled places, respectively. The corresponding relationship is determined by the experts' scoring based on their prior knowledge.

[0101] Optionally, the data after the dimensionality reduction process is clustered by a K-Means clustering algorithm to obtain clustered data divided into three clusters:

[0102] (1) Let t = 0 and randomly select 3 sample points as the initial cluster centers

[0103] (2) For a fixed cluster center in, For class G i The center of the sample X′ is calculated ij The distance to the cluster center is calculated as Assign each sample to the class with the closest center to form the clustering result C (t) ;

[0104] (3) Clustering result C (t) , calculate the mean of the samples in each current class as the new class center

[0105] (4) If the iteration converges or meets the stopping condition, output C * =C (t) ,get Three classes, otherwise t=t+1, return to step (2) and execute in a loop.

[0106] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A risk assessment and early warning method based on multi-dimensional information fusion, characterized in that: The risk assessment and early warning method comprises: Collect data of various dimensions of personnel and equipment in the scheduled place, wherein the data includes basic information data of preset personnel, scoring and evaluation data, danger level data of preset personnel, behavioral benchmark data of preset personnel, number of blacklist and whitelist calls in the scheduled place, laser perimeter alarm data, security system alarm data, and video surveillance data; Perform fusion preprocessing on the collected data, wherein the fusion preprocessing includes data filling, data cleaning, T-SNE dimensionality reduction processing, and K-Means clustering processing; Combining expert experience to build a priori knowledge base for risk assessment of scheduled places, and evaluating and predicting the risk level of scheduled places based on the priori knowledge base and the fusion preprocessed data, wherein the risk level of the scheduled places includes high risk, medium risk and low risk; The fusion preprocessing of the collected data is specifically as follows: After collecting business data of all dimensions, fill and clean the data each time, and then store it in the database; Perform T-SNE dimensionality reduction on the stored data, and reduce our nonlinear related data to 2 dimensions through the T-SNE dimensionality reduction algorithm; The data after dimensionality reduction is clustered using the K-Means clustering algorithm to obtain clustered data divided into three clusters, where the three clusters correspond to high, medium, and low risks of the scheduled places, respectively. The corresponding relationship is determined by the experts' scoring based on their prior knowledge. The data after dimensionality reduction is clustered by K-Means clustering algorithm, and the clustered data divided into three clusters are obtained as follows: (1) Let t = 0 and randomly select 3 sample points as the initial cluster centers (2) For a fixed cluster center in, For class G i The center of the sample X′ is calculated ij The distance to the cluster center is calculated as Assign each sample to the class with the closest center to form the clustering result C (t) ; (3) Clustering result C (t) , calculate the mean of the samples in each current class as the new class center (4) If the iteration converges or meets the stopping condition, output C * =C (t) ,get Three classes, otherwise t=t+1, return to step (2) and execute in a loop.

2. The risk assessment and early warning method according to claim 1 is characterized in that: The evaluation and prediction of the risk level of a predetermined place based on the prior knowledge base and the fusion pre-processed data are specifically as follows: Combining the data collected at the current moment with the historical data of the previous N days in the prior knowledge base, the data of various topics in the future period are predicted through the long short-term memory artificial neural network; Input the historical data of each dimension into the time series LSTM model to predict the data of different dimensions in the future, and obtain the possible data of the future in the same format as the original data; Based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, the risk level at the current moment is evaluated and the future risk level of the scheduled place is predicted.

3. The risk assessment and early warning method according to claim 2 is characterized in that: The method of evaluating the risk level at the current moment and predicting the future risk level of the scheduled place based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data in the future period of time is as follows: After obtaining the business data at the current moment, it is input into the prior knowledge base to obtain the risk results of the scheduled venue in the current scenario; Afterwards, the current business data and risk assessment results are input into the database. At the same time, combined with the current input data and the historical data of the previous N days in the database, the business data of each dimension in the future period is predicted separately through the long short-term memory artificial neural network, and the predicted data in the future period is input into the prior knowledge base to obtain the risk results of the scheduled venues in the future scenarios.

4. A risk assessment and early warning device based on multi-dimensional information fusion, characterized in that: The risk assessment and early warning device comprises: A collection unit is configured to collect data of various dimensions of personnel and equipment in a predetermined place, wherein the data includes basic information data of preset personnel, scoring and evaluation data, risk level data of preset personnel, behavioral benchmark data of preset personnel, number of blacklist and whitelist calls in the predetermined place, laser perimeter alarm data, security system alarm data, and video surveillance data; A processing unit, configured to perform fusion preprocessing on the collected data, wherein the fusion preprocessing includes data filling, data cleaning, T-SNE dimensionality reduction processing, and K-Means clustering processing; An evaluation and early warning unit is configured to construct a priori knowledge base for risk evaluation of a predetermined place in combination with expert experience, and to evaluate and predict the risk level of the predetermined place based on the priori knowledge base and the data after fusion preprocessing, wherein the risk level of the predetermined place includes high risk, medium risk and low risk; The processing unit is specifically configured as follows: After collecting business data of all dimensions, fill and clean the data each time, and then store it in the database; Perform T-SNE dimensionality reduction on the stored data, and reduce our nonlinear related data to 2 dimensions through the T-SNE dimensionality reduction algorithm; The data after dimensionality reduction is clustered using the K-Means clustering algorithm to obtain clustered data divided into three clusters, where the three clusters correspond to high, medium, and low risks of the scheduled places, respectively. The corresponding relationship is determined by the experts' scoring based on their prior knowledge. The data after dimensionality reduction is clustered by K-Means clustering algorithm, and the clustered data divided into three clusters are obtained as follows: (1) Let t = 0 and randomly select 3 sample points as the initial cluster centers (2) For a fixed class center in, For class G i The center of the sample X′ is calculated ij The distance to the cluster center is calculated as Assign each sample to the class with the closest center to form the clustering result C (t) ; (3) Clustering result C (t) , calculate the mean of the samples in each current class as the new class center (4) If the iteration converges or meets the stopping condition, output C * = C(t), we get Three classes, otherwise t=t+1, return to step (2) and execute in a loop.

5. The risk assessment and early warning device according to claim 4 is characterized in that: The evaluation and early warning unit is specifically configured as follows: Combining the data collected at the current moment with the historical data of the previous N days in the prior knowledge base, the data of various topics in the future period are predicted through the long short-term memory artificial neural network; Input the historical data of each dimension into the time series LSTM model to predict the data of different dimensions in the future, and obtain the possible data of the future in the same format as the original data; Based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data for a period of time in the future, the risk level at the current moment is evaluated and the future risk level of the scheduled place is predicted.

6. The risk assessment and early warning device according to claim 5 is characterized in that: The method of evaluating the risk level at the current moment and predicting the future risk level of the scheduled place based on the prior knowledge base, the current data after fusion preprocessing, and the predicted data in the future period of time is as follows: After obtaining the business data at the current moment, it is input into the prior knowledge base to obtain the risk results of the scheduled venue in the current scenario; Afterwards, the current business data and risk assessment results are input into the database. At the same time, combined with the current input data and the historical data of the previous N days in the database, the business data of each dimension in the future period is predicted separately through the long short-term memory artificial neural network, and the predicted data in the future period is input into the prior knowledge base to obtain the risk results of the scheduled venues in the future scenarios.

Images