A dangerous goods concealment monitoring system and method based on artificial intelligence

Abstract

The present invention discloses a dangerous goods concealment monitoring system based on artificial intelligence, including a data storage end, a data processing end, a model building end, an information reading end, a result display end, and a system management end; the model building end uses the declared information data as input data and the concealment detection result as output data, uses a training set to establish a cost-sensitive adaptive enhancement algorithm, and tests the model performance on a test set; the information reading end reads the declared information data of all container bookings in this batch; the result display end displays the dangerous goods concealment prediction results of all container bookings in this batch obtained by using the model building end algorithm. The method of the present invention realizes rapid and high-accuracy identification of high-risk bookings by utilizing a cost-sensitive adaptive enhancement algorithm to mine the association between the concealment of dangerous goods and the key attributes of the booking declaration information, thereby improving the efficiency of random inspections of dangerous goods concealment by transportation regulatory authorities.

Description

Technical Field

[0001] The present invention relates to a system and method for monitoring the concealment of dangerous goods, and in particular to a system and method for monitoring the concealment of dangerous goods based on artificial intelligence. Background Art

[0002] Dangerous goods are often transported in closed containers. Because the transportation cost of dangerous goods is significantly higher than that of ordinary cargo, some shippers maliciously conceal dangerous goods in their container declarations, creating potential risks of dangerous goods accidents and safety hazards, resulting in serious problems such as casualties, cargo damage, and environmental pollution. Despite increased spot checks by transportation regulators to address this issue, the current manual shorthand spot checks are inefficient, with only approximately 1% of bookings being found to contain undeclared dangerous goods. Therefore, the development of an AI-based dangerous goods undeclaration detection system and method can effectively select high-risk bookings with greater detection value, saving manpower and resources and improving spot checks efficiency.

[0003] Historical container booking data is crucial for analyzing the under-declaration of dangerous goods. Existing booking systems accumulate a vast amount of historical declaration data, including random inspection results and booking information for under-declared dangerous goods, but its value remains largely untapped. Booking information includes information such as booking date, cargo name, frequency and rating of shipper cooperation, frequency and rating of freight forwarder cooperation, frequency and rating of consignee cooperation, trade zone, and import / export terms. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide an artificial intelligence-based dangerous goods concealment monitoring system and method that can achieve rapid and highly accurate identification of high-risk bookings and improve the efficiency of random inspections of dangerous goods concealment by transportation regulatory authorities.

[0005] Technical solution: The dangerous goods concealment monitoring system of the present invention includes:

[0006] Data storage end, data processing end, model building end, information reading end, result display end and system management end;

[0007] The data storage end is used to store historical container booking data that has undergone dangerous goods concealment monitoring procedures;

[0008] The data processing end calls the data from the data storage end, performs data preprocessing on it, and divides the data into training sets and test sets;

[0009] On the model building side, the declared information data is used as input data, and the concealment detection results are used as output data. The training set is used to establish a cost-sensitive adaptive enhancement algorithm, and the model performance is tested on the test set.

[0010] The information reading terminal reads the declared information data of all container bookings in this batch;

[0011] The result display side uses the cost-sensitive adaptive enhancement algorithm based on the model building side to obtain the prediction results of dangerous goods concealment for all container bookings in this batch. The monitoring department then detects high-risk bookings that are judged to have dangerous goods concealment.

[0012] The system management terminal is used to update and manage data of the monitoring system.

[0013] Furthermore, the data storage terminal stores historical container booking data that has undergone dangerous goods concealment monitoring procedures, including concealment detection results and booking time, cargo name, shipper cooperation frequency and level, freight forwarder cooperation frequency and level, consignee cooperation frequency and level, trade zone and import and export terms.

[0014] Furthermore, the data processing end calls the data of the data storage end, deletes the historical container booking data containing missing values, and divides the remaining data into a training set and a test set after deleting the missing values.

[0015] The method for monitoring the concealment of dangerous goods of the present invention comprises the following steps:

[0016] S1, collects historical container booking data that has undergone dangerous goods concealment monitoring procedures through the data storage terminal;

[0017] S2, preprocesses the data collected by the data storage end in S1 through the data processing end, and divides the processed data into a training set and a test set;

[0018] S3 inputs the training set data into the cost-sensitive adaptive enhancement algorithm in the model building end, and uses the cost-sensitive adaptive enhancement algorithm to determine whether the container booking contains concealed dangerous goods.

[0019] Furthermore, in step S3, the steps for implementing the cost-sensitive adaptive enhancement algorithm are as follows:

[0020] Use the training set for training, input the training sample set Γ, the training sample set Γ contains N training samples, and the variable information of the nth training sample is (x n ,y n ), where y n ∈{-1,1} represents the concealed dangerous goods detection result of sample n, where -1 means that sample n does not contain concealed dangerous goods, and 1 means that sample n contains concealed dangerous goods; K characteristic variable values ​​representing the information parameters declared by sample n;

[0021] Construct M base classifiers, using random forest classifiers as the base classifier. The base classifier has three hyperparameters: the maximum number of features, the maximum depth, and the minimum number of samples in a leaf node. The optimal values ​​of the three hyperparameters are obtained through enumeration. The optimal value of the cost-sensitive factor η is also determined through enumeration. Indicates the sample weight of the nth training sample in the mth base classifier; assigns sample weights to each training sample in the training sample set, and the initialization value of the nth training sample in the first base classifier is Calculate the weight of the nth training sample of the base classifier m And calculate the cost sensitivity coefficient α of the base classifier m m , the specific steps are as follows:

[0022] S31, set m=1;

[0023] S32, using the training sample set Γ and sample weights Construct base classifier g m , its loss function L m for:

[0024]

[0025] Where g m (x n ) represents the base classifier g m The classification results of training sample n; α m represents the base classifier g m Cost sensitivity coefficient;

[0026] S33, using g m Classify the training sample set Γ and calculate the classification result of training sample n as g m (x n ), calculate the base classifier g m The error rate err m for:

[0027]

[0028] S34, calculate the base classifier g m The cost sensitivity coefficient α m for:

[0029]

[0030] S35, calculate the nth training sample in the base classifier g m+1 The sample weight for:

[0031]

[0032] S36, let m=m+1; if m≤M-1, return to step S3); otherwise, go to step S37;

[0033] S37, output the final classifier expression

[0034] when When H(x n ) value is +1; otherwise, H(x n ) value is -1.

[0035] Compared with the prior art, the present invention has the following significant effects:

[0036] 1. This invention uses a cost-sensitive adaptive enhancement algorithm to mine the correlation between the key attributes of dangerous goods misdeclaration and booking declaration information through container booking information, achieving rapid and highly accurate identification of high-risk bookings and improving the efficiency of random inspections of dangerous goods misdeclaration by transportation regulatory authorities.

[0037] 2. This invention collects historical container booking data that has undergone dangerous goods concealment monitoring procedures as input to the model building end, focusing on risk control at the source of transportation, and has practical significance in the field of dangerous goods transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural diagram of the monitoring system of the present invention;

[0039] Figure 2 Flow chart of the monitoring method of the present invention. DETAILED DESCRIPTION

[0040] The following describes the technical solutions of the artificial intelligence-based dangerous goods concealment monitoring system and method in the embodiments of the present invention in conjunction with the accompanying drawings. The described embodiments are only part of the embodiments of the present invention, not all of them.

[0041] The present invention is further described in detail using the container booking data of Ningbo Shipping that has undergone dangerous goods concealment monitoring procedures from 2012 to 2018, in conjunction with the accompanying drawings and specific implementation methods.

[0042] like Figure 1 As shown, the dangerous goods concealment monitoring system of the present invention includes:

[0043] Data storage end, data processing end, model building end, information reading end, result display end and system management end;

[0044] The data storage terminal is used to store historical container booking data of Ningbo Shipping that has undergone dangerous goods concealment monitoring procedures from 2012 to 2018. Each booking data includes one dangerous goods concealment inspection result and 12 declaration information parameters, including order month, contract agreement, export terms, import terms, gate terms, electronic booking, trade route, cargo category, frequency of shipper cooperation, frequency of freight forwarder cooperation, shipper cooperation level, and freight forwarder cooperation level.

[0045] On the data processing side, the data from the data storage side is called for data preprocessing, and the data is randomly divided into a training set and a test set in a ratio of 7:3;

[0046] On the model building side, the declared information data is used as input data, and the concealment detection results are used as output data. The training set is used to establish a cost-sensitive adaptive enhancement algorithm, and the model performance is tested on the test set.

[0047] The information reading terminal reads the 12 declared information data of all container bookings in this batch;

[0048] The result display side uses the cost-sensitive adaptive enhancement algorithm based on the model building side to obtain the prediction results of dangerous goods concealment for all container bookings in this batch. The monitoring department then detects high-risk bookings that are judged to have dangerous goods concealment.

[0049] On the system management side, the declaration information and test results of the container bookings for this batch of random inspections are entered into the historical data set, and the monitoring system data is updated.

[0050] like Figure 2 As shown, the method for monitoring the concealment of dangerous goods of the present invention comprises the following steps:

[0051] S1 collects historical container booking data that has undergone dangerous goods misdeclaration monitoring procedures through data storage. This data includes 300,000 booking records from Ningbo Shipping. Each booking record includes one dangerous goods misdeclaration inspection result and 12 information parameters, including order month, contract agreement, export terms, import terms, gate terms, electronic booking, trade route, cargo category, frequency of shipper cooperation, frequency of freight forwarder cooperation, shipper cooperation level, and freight forwarder cooperation level.

[0052] S2, preprocesses the data collected in S1 through the data processing end, deletes the historical container booking data containing missing values, and after deleting the missing values, randomly divides the remaining data into a training set and a test set in a ratio of 7:3.

[0053] S3 inputs the training set data from the data processing end in S2 into the cost-sensitive adaptive enhancement algorithm in the model building end, and uses the cost-sensitive adaptive enhancement algorithm to determine whether the container booking contains concealed dangerous goods.

[0054] Training using the training set: Input the training sample set Γ, Γ contains N training samples, and the variable information of the nth training sample is (x n ,y n ), where y n ∈{-1,1} represents the concealed dangerous goods detection result of sample n (-1 means that sample n does not contain concealed dangerous goods; 1 means that sample n contains concealed dangerous goods). K = 12 characteristic variable values ​​representing the declared information parameters of sample n; M = 100 base classifiers are constructed, and the base classifier is a random forest classifier. The base classifier contains three hyperparameters, namely the maximum number of features (with a value range of [3, 8]), the maximum depth (with a value range of [5, 11]), and the minimum number of samples contained in the leaf node (with a value range of [1, 3]). The optimal values ​​of the three hyperparameters are obtained by enumeration method to be 8, 10, and 1, respectively; the optimal value of the cost sensitivity factor η is obtained by enumeration method to be 0.01 (with a value range of η∈(0, 1)); Indicates the sample weight of the nth training sample in the mth base classifier; assigns sample weights to each training sample in the training sample set, and the initialization value of the nth training sample in the first base classifier is

[0055] Calculate the weight of the nth training sample of the base classifier m And calculate the cost sensitivity coefficient α of the base classifier m m , the specific steps are as follows:

[0056] Step 1) Set m=1;

[0057] Step 2) Using the training sample set Γ and sample weights Construct base classifier g m , its loss function L m for:

[0058]

[0059] Where g m (x n ) represents the base classifier g m The classification results of training sample n; α m represents the base classifier g m Cost sensitivity coefficient;

[0060] Step 3) Using g m Classify the training sample set Γ and calculate the classification result of training sample n as g m (x n ), calculate the base classifier g m The error rate err m for:

[0061]

[0062] Step 4) Calculate the base classifier g m The cost sensitivity coefficient α m for:

[0063]

[0064] Step 5) Calculate the nth training sample in the base classifier g m+1 The sample weight for:

[0065]

[0066] Step 6) Set m=m+1; if m≤M-1, return to step 2); otherwise, go to step 7);

[0067] Step 7) Output the final classifier expression in style hour, The value is +1, otherwise its value is -1.

[0068] The model performance was tested on the test set, and the test results were shown in Table 1.

[0069] Table 1 Model performance on the test set

[0070]

[0071] Table 1 shows that the cost-sensitive adaptive boosting algorithm performs well when processing an extremely unbalanced sample of approximately 1% of bookings containing undeclared dangerous goods. It can fully exploit the correlation between undeclared dangerous goods and the key attributes of booking declaration information, enabling rapid and highly accurate identification of high-risk bookings and improving the efficiency of random inspections of undeclared dangerous goods by transportation regulatory authorities.

[0072] The above embodiments should be understood as merely illustrating the present invention and not as limiting the scope of protection of the present invention. Those skilled in the art may make various changes or modifications to the present invention based on the concept of the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A method for monitoring the concealment of dangerous goods based on artificial intelligence, characterized in that: The following steps are involved: S1, collects historical container booking data that has undergone dangerous goods concealment monitoring procedures through the data storage terminal; S2, preprocesses the data collected by the data storage end in S1 through the data processing end, and divides the processed data into a training set and a test set; S3: Input the training set data into the cost-sensitive adaptive boosting algorithm in the model building end. This algorithm is used to determine whether the container booking contains undeclared dangerous goods. In step S3, the steps for implementing the cost-sensitive adaptive enhancement algorithm are as follows: Use the training set for training, input the training sample set Γ, the training sample set Γ contains N training samples, and the variable information of the nth training sample is (x n ,y n ), where y n ∈{-1,1} represents the concealed dangerous goods detection result of sample n, where -1 means that sample n does not contain concealed dangerous goods, and 1 means that sample n contains concealed dangerous goods; K characteristic variable values ​​representing the information parameters declared by sample n; Construct M base classifiers, using random forest classifiers as the base classifier. The base classifier has three hyperparameters: the maximum number of features, the maximum depth, and the minimum number of samples in a leaf node. The optimal values ​​of the three hyperparameters are obtained through enumeration. The optimal value of the cost-sensitive factor η is also determined through enumeration. Indicates the sample weight of the nth training sample in the mth base classifier; assigns sample weights to each training sample in the training sample set, and the initialization value of the nth training sample in the first base classifier is Calculate the weight of the nth training sample of the base classifier m And calculate the cost sensitivity coefficient α of the base classifier m m , the specific steps are as follows: S31, set m=1; S32, using the training sample set Γ and sample weights Construct base classifier g m , its loss function L m for: Where g m (x n ) represents the base classifier g m The classification results of training sample n; α m represents the base classifier g m Cost sensitivity coefficient; S33, using g m Classify the training sample set Γ and calculate the classification result of training sample n as g m (x n ), calculate the base classifier g m The error rate err m for: S34, calculate the base classifier g m The cost sensitivity coefficient α m for: S35, calculate the nth training sample in the base classifier g m+1 The sample weight for: S36, let m=m+1; if m≤M-1, return to step S3); otherwise, go to step S37; S37, output the final classifier expression when When H(x n ) value is +1; otherwise, H(x n ) value is -1.

2. An artificial intelligence-based dangerous goods concealment monitoring system for executing the dangerous goods concealment monitoring method according to claim 1, characterized in that: include: Data storage end, data processing end, model building end, information reading end, result display end and system management end; The data storage end is used to store historical container booking data that has undergone dangerous goods concealment monitoring procedures; The data processing end calls the data from the data storage end, performs data preprocessing on it, and divides the data into training sets and test sets; On the model building side, the declared information data is used as input data, and the concealment detection results are used as output data. The training set is used to establish a cost-sensitive adaptive enhancement algorithm, and the model performance is tested on the test set. The information reading terminal reads the declared information data of all container bookings in this batch; The result display side uses the cost-sensitive adaptive enhancement algorithm based on the model building side to obtain the prediction results of dangerous goods concealment for all container bookings in this batch. The monitoring department then detects high-risk bookings that are judged to have dangerous goods concealment. The system management terminal is used to update and manage data of the monitoring system.

3. The artificial intelligence-based dangerous goods concealment monitoring system according to claim 2 is characterized by: The data storage terminal stores historical container booking data that has undergone dangerous goods concealment monitoring procedures, including concealment detection results and booking time, cargo name, consignor cooperation frequency and level, freight forwarder cooperation frequency and level, consignee cooperation frequency and level, trade zone and import and export terms.

4. The artificial intelligence-based dangerous goods concealment monitoring system according to claim 2 is characterized by: The data processing end calls the data of the data storage end, deletes the historical container booking data containing missing values, and divides the remaining data into a training set and a test set after deleting the missing values.

Images