TMR-4930-Ship-trajectory-prediction-in-confined-waters

TMR 4930 Ship trajectory prediction in confined waters

AIS dataset process files

1) 0_ais_dataset_process_hs-

• Dataset split by months
• Visualization:
  • By MMSI population
  • By message time interval
  • Map plots

2) 1_ais_dataset_process_hs-

• Dataset processing:
  • Handling column names
  • Voyage identification

3) 2_ais_dataset_process_hs_uniform_ts

• Dataset processing:
  • By navigation status
  • Uniform timestamp generation
  • Handling features:
    • Distance interpolation
    • Speed over ground (SOG) calculation
    • Course over ground (COG) interpolation
    • Handling voyage halts/zero SOG scenarios (sub voyage identification)
• Visualizations:
  • By navigation status
  • By lower SOG values (slow vessel dynamics)

4) 3_ais_dataset_process_hs_news_uniform_ts

(Note: NEWS has not been tried as a feature for model training as a part of this work.)

• Dataset processing/feature engineering:
  • NEWS generation (distance to coast towards North, East, West, and South direction) - geolocation generalized feature
  • Other feature engineering (* have been used as a part of training data):
    • The Vessel is in TSS (In_TSS)
    • The vessel belongs to TSS (Vessel_TSS)
    • Nearest distance to coast (NDC)

5) 4_ais_dataset_process_hs_ttv_rnn

• Dataset processing:
  • Standardization (Min-Max)
• Dataset creation:
  • Create sequences of data with a length of 20 timestamps.
  • For the nth timestamp in a voyage, the sequence [n:n+20) is created, where [n:n+10) represents the observed trajectory (X) and [n+10:n+20) represents the predicted trajectory (Y).

6) 5_ais_dataset_process_hs_other_encounter

• Dataset processing:
  • Used to consider all vessels towards encounter vessels, irrespective of zero SOG/stationary status.

7) 6_ais_dataset_process_hs

• Training dataset creation:
  • Concatenate data from months January to October to create the training dataset.
  • Separate data into:
    • X_train: Input features
    • Y_lat_train: Ground truth latitude for training
    • Y_lon_train: Ground truth longitude for training
    • Y: Combined ground truth latitude and longitude for training

8) 7_ais_dataset_process_encounter_uniform_ts

(Note: ENCOUNTERS have not been used as a part of the training/this work)

• Dataset generator:
  • Identification of encounter vessels for each dynamic vessel (Encounter vessel data generator).
  • Accommodating encounters could improve predictions and provide real-world scenario representation.

9) 8_ais_dataset_process_dist_angle

• Dataset generator:
  • Implementation of the proposed RDA (Relative Displacement Angle) approach.
  • Map the relative displacement of prediction timestamps to the last observed state of the 10-minute voyage set.
  • Map the relative change in the bearing angle to the last observed state of the 10-minute voyage set.

ml models

- 3 models developed towards prediction and comparison:

1) CNN LSTM

2) LSTM

3) GRU

- Training/predictions made using two approaches:

1) Normal - latitude, longitude predictions (file tags: _normal)

2) RDA - displacement/distance, bearing/angle (file tags: _with_distance, _with_angle, _rda_model)