A ship traffic flow prediction method based on missing information perception
By constructing missing features and feature extraction models, and combining them with ensemble learning methods, the problem of reduced prediction accuracy caused by missing AIS data was solved, and more accurate ship traffic flow prediction was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUILIN UNIVERSITY OF TECHNOLOGY
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-05
AI Technical Summary
Existing ship traffic flow prediction methods suffer from reduced accuracy when AIS data is missing, and traditional processing methods may destroy the structural characteristics of the original time series data.
We construct missing mask features and missing time interval features, combine them with the Transformer model for feature extraction, and fuse the results of multiple prediction models through the AdaBoost ensemble learning method to achieve effective utilization of missing data.
It improves the accuracy and stability of predictions in the case of missing data, and can more accurately predict ship traffic flow.
Smart Images
Figure CN122157522A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent shipping and traffic information processing technology, specifically to a method for predicting ship traffic flow based on Automatic Identification System (AIS) data, and more particularly to a method for predicting ship traffic flow that integrates missing information modeling and deep learning prediction. Background Technology
[0002] With the rapid development of the global shipping industry, changes in ship traffic flow in waterways have a significant impact on shipping scheduling and management, waterway planning, and shipping safety. Accurate prediction of ship traffic flow can provide decision-making support for waterway management departments, thereby improving shipping efficiency and reducing navigation risks.
[0003] Automatic Identification Systems (AIS) can collect information such as a ship's position, time, and navigation status in real time, providing an important data source for ship traffic flow prediction. Currently, AIS data is widely used in shipping traffic analysis, track prediction, and ship behavior research.
[0004] However, in actual AIS data acquisition, due to factors such as complex communication environments, equipment failures, and signal obstruction, AIS data often suffers from problems such as missing data and uneven sampling time. These problems lead to incomplete time-series data, thus affecting the prediction accuracy of ship traffic flow forecasting models.
[0005] Existing traffic flow forecasting methods typically assume that the input data is complete. For missing data, they often employ methods such as deletion or interpolation. However, these methods may disrupt the structural characteristics of the original time series data, leading to a decline in model prediction accuracy. Therefore, how to effectively utilize AIS data and improve the accuracy of ship traffic flow forecasting in the presence of missing data has become a crucial technical problem that needs to be solved in the field of shipping traffic data analysis. Summary of the Invention
[0006] The purpose of this invention is to provide a method for predicting ship traffic flow based on missing information perception, so as to solve the problem of reduced prediction accuracy when AIS data is missing.
[0007] To achieve the above objectives, the present invention adopts the following technical solution:
[0008] A method for predicting ship traffic flow based on missing information perception includes the following steps:
[0009] S1, acquire Automatic Identification System (AIS) data, and perform data cleaning and time alignment processing on the AIS data.
[0010] S2 aggregates AIS data over time according to a preset time interval, counts the number of ships per unit time, and thus constructs a time series of ship traffic flow.
[0011] S3, perform missing detection on the ship traffic flow time series, and construct missing mask features and missing time interval features to represent the data missing status in the time series.
[0012] S4, construct an input feature sequence, including historical traffic features, time features, and differential features, and add the missing mask features and missing time interval features to the input feature sequence.
[0013] S5 uses the Transformer model to extract time-series features from the input feature sequence, thereby obtaining the initial prediction results of ship traffic flow.
[0014] S6 uses the AdaBoost ensemble learning method to weight and fuse the outputs of multiple prediction models to obtain the final ship traffic flow prediction result.
[0015] Compared with the prior art, the present invention has the following advantages:
[0016] (1) By constructing missing mask features and missing time interval features, the missing information in AIS data is effectively represented, which improves the model's adaptability to missing data situations;
[0017] (2) Using the Transformer model to extract features from time series data can better capture the temporal characteristics of changes in ship traffic flow;
[0018] (3) The results of multiple prediction models are fused by using the AdaBoost ensemble learning method, which improves the stability and accuracy of the prediction results.
[0019] Therefore, this invention can achieve more accurate ship traffic flow prediction in the presence of missing data, and can be applied to fields such as waterway traffic monitoring, shipping scheduling management, and intelligent shipping decision support. Attached Figure Description
[0020] Figure 1 This is a flowchart of the overall process for predicting ship traffic flow proposed in this invention.
[0021] Figure 2 Flowchart for constructing the missing information sensing features proposed in this invention;
[0022] Figure 3 This is a structural diagram of a ship traffic flow prediction model based on Transformer and AdaBoost. Detailed Implementation
[0023] To make the technical solution of the present invention clearer, the present invention will be further described below with reference to the accompanying drawings.
[0024] First, the raw data from the Automatic Identification System (AIS) is acquired. This AIS data includes information such as the ship's identifier, longitude, latitude, and timestamp. The acquired AIS data then undergoes data cleaning processing, including deleting abnormal data, removing duplicate data, and standardizing the time format.
[0025] The cleaned AIS data is then time-aligned and aggregated according to preset time intervals. For example, AIS data can be statistically analyzed at fixed time intervals to obtain the number of ships per unit time and construct a time series of ship traffic flow.
[0026] After constructing the time series, missing data detection is performed. If no ship data is detected at a certain time point, it is considered that data is missing at that time point. A missing data mask feature is constructed to indicate whether observation data exists at the corresponding time point. Further, the missing data mask feature is defined as: M t =1 indicates that observational data exists at this time point; M t = 0 indicates that there is no observation data at this time point. Meanwhile, the missing time interval feature is defined as: D t This represents the time interval between the current time point and the time point where the most recent observation data existed.
[0027] Based on this, an input feature sequence is constructed. The input features include historical traffic flow features, time features, and difference features. Historical traffic flow features represent changes in ship traffic flow within past time windows, time features represent time period information, and difference features describe traffic flow trends. These features are then fused with missing mask features and missing time interval features to form the model input features.
[0028] Subsequently, the Transformer model is used to extract temporal features from the input feature sequence. The Transformer model, through its self-attention mechanism, can capture the dependencies between different time points in the time series, thereby improving the model's ability to learn the changing patterns of ship traffic flow.
[0029] Finally, the outputs of multiple prediction models are weighted and fused using the AdaBoost ensemble learning method. This method assigns different weights to each model based on their prediction errors, thereby obtaining the final ship traffic flow prediction result.
[0030] The above methods can achieve more accurate ship traffic flow prediction even when AIS data is missing, thus providing effective data support for waterway traffic management and shipping decisions.
Claims
1. A method for predicting ship traffic flow based on missing information perception, characterized in that, Includes the following steps: S1. Acquire Automatic Identification System (AIS) data and perform data cleaning and time alignment processing; S2. Perform time aggregation on AIS data according to preset time intervals, and count the number of ships per unit time to construct a time series of ship traffic flow. S3. Perform missing detection on the time series and construct missing mask features and missing time interval features; S4. Construct an input feature sequence, which includes historical traffic features, time features, and differential features, and fuse the missing mask features and missing time interval features into the input feature sequence; S5. Use the Transformer model to extract temporal features from the input feature sequence to obtain the initial prediction result; S6. The AdaBoost ensemble learning method is used to weight and fuse multiple prediction results to obtain the ship traffic flow prediction result.
2. The method according to claim 1, characterized in that, The AIS data includes the ship's identifier, longitude, latitude, timestamp, and navigation status information.
3. The method according to claim 1, characterized in that, The time alignment process involves uniformly sampling AIS data at fixed time intervals to construct time-series data with equal time intervals.
4. The method according to claim 1, characterized in that, The missing mask feature is a binary variable used to represent the data existence status at the corresponding time point in the time series.
5. The method according to claim 1, characterized in that, The missing time interval feature is used to represent the time difference between the current time point and the most recent valid observation time point.
6. The method according to claim 1, characterized in that, The Transformer model includes a multi-head self-attention mechanism and a feedforward neural network structure, which are used to extract temporal dependencies in time series.
7. The method according to claim 1, characterized in that, The AdaBoost ensemble learning method achieves weighted fusion of prediction results by evaluating and assigning weights to the prediction errors of multiple prediction models.
8. The method according to claim 1, characterized in that, The ship traffic flow prediction results are used for waterway traffic monitoring, shipping scheduling management, and intelligent shipping decision support.