Vessel - Feature

[pagand]

Dynamic_real_feature

• current = STW - SOG
• SPEED = average of speed_1 & 2
• Mode
• turn = Heading - previous Heading
• acceleration
• distance
• rain/snowfall/wind_force/wind_dirc
• resist_ratio (avg of ratio_1 & 2)
• disp_x = (changes in x-axis calculated by heading angle) * sog
• disp_y = (changes in y-axis calculated by heading angle) * sog

Static_Categorical_feature

• is_weekday
• direction
• ScheduleType
• season
• hour

Input

• FC2 = FC^2
• SOG4 = SOG^4
• LATITUDE
• LONGITUDE

Time_feature: countDown

[pagand]

@tzhao-ooc Looks good. some thing to notice: 1- put the time feature bellow the other Dynamic_real_feature if they are same input to avoid confusion. 2- Change FC^2, SOG^4 to their original space if it is between 0 and 1. if not, you can try to make them between 0,1 but smaller is not suggested. 3- change the input title to previous input values 4- for the dis_x, dis_y make sure the sog is for previous time step not the current/future ones 5- find the list of actions (manipulative variables). These are the one as input for our RL method. You can consider them as a part of Dynamic_real_feature. The only difference is that, the others are either disturbances (out of our control) or engineered feature that we think based on the domain knowledge. Understand these concept in the paper 2 to help you which features should be included which not.

6- find out if there is a separate input in the informer for "Input" which is our previous output variable. or are they added to Dynamic_real_feature?

Important. Put the dates in your issue. The goal is to reach R^2>0.95 for all the features that we are computing. You can change the features, change the model, combine the output (one model for all 4) or any other edit to reach this. My only concern is timing. We need to achieve this by end of June.

[pagand]

@tzhao-ooc Please update your status and let us know what you are working on. Here is a suggested breakdown of the different features:

Static states: trip start hour, direction​

Dynamic states: heading rate, resistance (torque/thrust), displacement, previous outputs, and disturbance-related variables​

Disturbances: time, weekday, current, season, weather, wind direction, wind force, and water depth​

Outputs: latitude, longitude, SOG, and fuel consumption​

Actions: heading, shaft speed, and mode selections​

Observations: current and next states

[tzhao-ooc]

I'm still working on modifying dynamic features to see the result of the model. I’m trying with different features and using two models or combine to one model.