Machine learning-based north pole ship accident severity prediction method
Through machine learning algorithms combining multi-source data and human-cause analysis framework, we predict the severity of Arctic ship accidents, solving the problem of traditional methods relying on a single data source and ignoring human factors, and improving prediction accuracy and comprehensiveness.
Patent Information
- Application Number
- CN202411201599.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-31
AI Technical Summary
The traditional Arctic ship accident prediction method relies on a single data source, cannot make full use of multi-source data, and lacks consideration of human factors, resulting in low prediction accuracy and difficulty in preventing accidents in real time.
Using machine learning algorithms, combining ship accident data, meteorological and hydrological information and human factors, through random forest model and human-cause analysis framework, predict the severity of accidents, supplement missing data, deal with data imbalance problem, and improve model robustness and generalization. ability.
It effectively improves the accuracy and comprehensiveness of the prediction of accident severity, can better reflect the background of complex accidents, considers human factors, and improves the accuracy and practicality of predictions.
Smart Images
Figure CN119227867A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of severity prediction of Arctic ship accidents, and in particular to a method for predicting severity of Arctic ship accidents driven by multi-source data using a machine learning algorithm. Background Art
[0002] The navigation conditions in the Arctic region are harsh, and accidents often occur in ships at sea in the Arctic due to factors such as sea ice, meteorological environment, and human errors. These accidents not only threaten life and property, but also cause serious damage to the environment. Traditional accident analysis and prevention methods are relatively backward, making it difficult to predict and prevent accidents in real time. Traditional methods for predicting maritime accidents mainly rely on a single data source for prediction, and fail to fully utilize the advantages of multi-source data, resulting in low prediction accuracy. The existing methods for predicting ship accidents mainly have the following problems: Single data source: Traditional methods mostly rely on a single data source, such as accident reports, meteorological data, etc., lack the comprehensive use of multi-source data, and cannot fully reflect the complex background and multiple influencing factors of the accident. Lack of consideration of human factors: Human factors are an important cause of accidents at sea, but traditional methods have little consideration of human factors, making it difficult to fully predict the severity of accidents. Limited prediction accuracy of the model: The prediction accuracy of traditional descriptive statistical analysis is limited, which is difficult to meet the needs of practical applications. Summary of the invention
[0003] In view of the problems existing in the prior art, the present invention discloses a method for predicting the severity of Arctic ship accidents based on machine learning, which specifically includes the following steps:
[0004] Obtain the data of ship accident lists in Arctic waters over the years based on the Global Ship Information System;
[0005] According to the time and geographical coordinates of the ship accident, the meteorological and hydrological information at the time of the ship accident was retrieved in ERA5, and the attribute information of the accident ship was supplemented based on the ship database;
[0006] Supplement some missing data in the list of ship accidents in the Global Ship Information System;
[0007] Based on the ship accident investigation report in the accident information provided by the Global Ship Information System, the human factor analysis and classification system framework is used to collect human information when the ship accident occurs;
[0008] The ship accident list data, accident ship attribute information and human information when the ship accident occurs are collected and divided into training set and test set. The training set is input into the random forest model to train the model. The accuracy, precision, recall rate and F1 score are used to evaluate the accuracy of the model in judging the severity of ship accidents. The test set is input into the trained random forest model, and the severity of the accident is predicted with the severity of the accident as the target value and the other variables as feature values.
[0009] Furthermore, the target value is a ship accident risk level, including not serious, serious and very serious.
[0010] Furthermore, the characteristic values include: accident type, accident month, accident time, location of the accident, gross tonnage of the ship, length, width of the ship, wind speed, temperature, air pressure, wave height, unsafe behavior, preconditions for unsafe behavior, unsafe supervision and organizational impact.
[0011] Furthermore, the accident information includes accident date, accident time, accident location, and accident type.
[0012] Furthermore, the ship information includes ship type, age, length, width and gross tonnage of the ship.
[0013] Furthermore, the meteorological and hydrological information includes wind speed, wave height, air pressure, temperature, water vapor evaporation and sea ice coverage information.
[0014] Due to the adoption of the above technical scheme, the present invention provides a method for predicting the severity of Arctic ship accidents based on machine learning. The method adopts the fusion of multi-source data (including ship accident data, meteorological and hydrological information, human factors, etc.), which effectively improves the accuracy and comprehensiveness of accident severity prediction. This fusion method can better reflect the complex scenario when the accident occurs, and improves the limitations of the single data source of the traditional method. In addition, the method takes human factors into consideration. The technical scheme incorporates human factors into the prediction of accident severity, analyzes the influencing factors such as operator behavior and organization when the accident occurs through the human factor analysis and classification framework (HFACS), and comprehensively evaluates the cause of the accident. The method also supplements missing data and handles sample imbalance problems. The invention introduces a random forest algorithm to fill in the missing data, so that the prediction model has better robustness and can handle the problem of incomplete data in reality. The synthetic minority class oversampling technology (SMOTE) is used to balance the data set to avoid the overfitting problem of the model due to insufficient minority class samples, thereby improving the generalization ability of the model. The technology uses a machine learning algorithm, especially a random forest algorithm, to predict the severity of Arctic ship accidents. Compared with traditional descriptive statistical analysis, machine learning models have higher prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0016] Figure 1 The flowchart of the method for predicting the severity of Arctic ship accidents based on machine learning in the method of the present invention.
[0017] Figure 2 The random forest model algorithm is used in an embodiment of the present invention to predict the contribution of different characteristics of the severity of Arctic ship accidents.
[0018] Figure 3 It is an individual conditional expectation diagram (ICE) of ship characteristics involved in the random forest algorithm to predict the severity of Arctic ship accidents in an embodiment of the present invention, which shows the dependence of the prediction on the characteristics of each instance.
[0019] Figure 4 FIG. 1 is a schematic diagram of the accuracy of accident severity prediction using different types of features in an embodiment of the present invention. FIG. DETAILED DESCRIPTION
[0020] In order to make the technical solutions and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention:
[0021] The severity of ship accidents at sea in the Arctic is affected by a variety of factors, including the type of accident, accident area, time, ship characteristics, meteorological factors, human factors, etc. The present invention combines the multiple factors affecting the severity of Arctic ship accidents, develops an Arctic ship accident database, and uses a machine learning algorithm to analyze the important factors affecting the severity of ship accidents. The method has application prospects in promoting safe navigation of ships in Arctic waters and predicting the severity of Arctic ship accidents in the future.
[0022] like Figure 1 As shown, the present invention provides a method for predicting the severity of Arctic ship accidents based on machine learning, which mainly includes the following steps:
[0023] S1. Retrieve the accident list: Retrieve the ship accidents north of 58°N from January 1, 2012 to December 31, 2022 from the GISIS (Global Integrated Shipping Information System) database. These include less serious accidents, serious accidents, very serious accidents, and the accident severity classification standard is based on MSC-MEPC.3 / Circ.4 / Rev.1 issued by IMO. The ship accident list provided by GISIS retains the flag state, gross tonnage, accident time, accident type, coordinate location, location, initializing event, summary of events, ship types, and consequence.
[0024] S2. Collect ship attribute information: The ship accident list provided by GISIS lacks ship attribute information or some ship type data is missing. According to the ship name and IMO number of the ship, query Balticshaipping.com to supplement the accident ship age (Age), length (Length Overall), ship width (Breadth Extreme), and some ship types (Ship Types) and gross tonnage (Gross Tonnage) missing in the list.
[0025] S3. Collect meteorological information: Use the coordinate information of the ship at the time of the accident and the time of the accident provided by GISIS to query the meteorological information at the time of the accident in ERA5. Supplementary data include wind speed, temperature, air pressure, wave height, sea ice coverage, and water vapor evaporation.
[0026] S4. Human factors in ship accidents: The Human Factor Analysis and Classification System (HFACS) framework is used to identify human factors in ship accidents from ship accident investigation reports and summary of events. Some ship accident investigation reports in the GISIS database are in the "Not available for download" status. To supplement the remaining accident investigation reports, according to the "Reporting Administration" provided by GISIS, the ship accident investigation agencies in relevant countries conduct cross-searches on the accident report release websites of relevant countries based on the ship's IMO number, ship name, and accident time to obtain ship accident investigation reports. The remaining accident investigation reports that could not be obtained were extracted from the "Summary of Events" of GISIS, and finally the accidents for which human factors in the accidents could not be obtained were eliminated.
[0027] S5: Missing data filling. Random forest is robust to missing values and can handle data sets containing missing values. Therefore, we choose to use the random forest algorithm in the sklearn library to fill in the missing value variables.
[0028] S6. Encode the category features as shown in Table 1.
[0029] Table 1
[0030]
[0031]
[0032] S7. To address the problem of data imbalance, the synthetic minority oversampling technique (SMOTE) is used. It is an improved solution based on the random oversampling algorithm. Since random oversampling increases minority samples by duplicating samples, it is easy to cause model overfitting problems. This will make the information learned by the model less generalized. The SMOTE algorithm solves this problem by analyzing minority samples and artificially synthesizing new samples to add to the data set. The following is a description of the algorithm process. First, for each sample X in the minority class, it is necessary to calculate the distance from it to all samples in the minority class sample set based on the Euclidean distance to obtain its k nearest neighbors; secondly, a sampling ratio is set according to the sample imbalance ratio to determine the sampling multiplier N. For each minority class sample X, a number of samples are randomly selected from its k nearest neighbors. Assume that the selected nearest neighbors are X n , for each randomly selected neighbor X n , and construct new samples with the original samples in the following ways.
[0033] X new=X+rand(0,1)*|XX n |
[0034] S7. Divide the data after equalization into a training set and a test set.
[0035] S8, use accuracy, precision, recall and F-1 score to evaluate the performance of different models. When the model predicts the positive class and the prediction is correct, it is called TruePositive; when the model predicts the negative class and the prediction is correct, it is called TrueNegative; when the model predicts the positive class, but the prediction is wrong, it is called False Positive; when the model predicts the negative class, but the prediction is wrong, it is called False Negative. The calculation formulas of the four indicators are as follows. The calculation results according to the calculation formulas of the indicators are shown in Table 2. By comparison, it can be seen that the Random Forest model has the best prediction performance.
[0036]
[0037]
[0038] Table 2
[0039]
[0040] S9. Use the Random Forest model to identify the main features that affect the severity of the accident. Figure 2 shown.
[0041] S10. In order to further explore the prediction differences of machine learning models with different category features, different category features are predicted using different machine learning models in this example. Figure 4 The results show that human factors perform well in machine learning applications, and the average prediction value of each model reaches 0.732.
[0042] Individual Conditional Expectation (ICE) plots show the dependence of predictions on the features of each instance. ICE plots play a key role in explaining machine learning models. ICE plots can help us understand how the model's predictions for a single sample change under different feature values. By visualizing the differences in individual predictions, we can better understand how the model makes personalized predictions based on the input features. Figure 3 As shown, Figure 3 The impact of ship characteristics on accident severity is demonstrated.
[0043] In the embodiment of the present invention, a framework of influencing factors affecting the severity of Arctic ship accidents is constructed, which is helpful to analyze and study various influencing factors affecting the severity of Arctic ship accidents from multiple perspectives.
[0044] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A method for predicting the severity of Arctic ship accidents based on machine learning, characterized in that include: Obtain the data of ship accident lists in Arctic waters over the years based on the Global Ship Information System; According to the time and geographical coordinates of the ship accident, the meteorological and hydrological information at the time of the ship accident was retrieved in ERA5, and the attribute information of the accident ship was supplemented based on the ship database; Supplement some missing data in the list of ship accidents in the Global Ship Information System; Based on the ship accident investigation report in the accident information provided by the Global Ship Information System, the human factor analysis and classification system framework is used to collect human information when the ship accident occurs; The ship accident list data, accident ship attribute information and human information when the ship accident occurs are collected and divided into training set and test set. The training set is input into the random forest model to train the model. The accuracy, precision, recall rate and F1 score are used to evaluate the accuracy of the model in judging the severity of ship accidents. The test set is input into the trained random forest model, and the severity of the accident is predicted with the severity of the accident as the target value and the other variables as feature values.
2. The method for predicting severity of Arctic ship accidents based on machine learning according to claim 1, characterized in that: The target value is the ship accident risk level, including not serious, serious and very serious.
3. The method for predicting severity of Arctic ship accidents based on machine learning according to claim 1, characterized in that: The characteristic values include: accident type, accident month, accident time, location of accident, gross tonnage of the ship, length, width of the ship, wind speed, temperature, air pressure, wave height, unsafe behavior, preconditions of unsafe behavior, unsafe supervision and organizational impact.
4. The method for predicting severity of Arctic ship accidents based on machine learning according to claim 1, characterized in that: The accident information includes accident date, accident time, accident location, and accident type.
5. The method for predicting severity of Arctic ship accidents based on machine learning according to claim 1, characterized in that: The ship information includes ship type, age, length, width and gross tonnage of the ship.
6. The method for predicting severity of Arctic ship accidents based on machine learning according to claim 1, characterized in that: The meteorological and hydrological information includes wind force, wave height, air pressure, temperature, water vapor evaporation and sea ice coverage information.