Brainstorming Additional Modeling Methods

1
Suggestion for Future Research on Parker Data
Brainstorming Additional Modeling Methods

• Apply reverse modeling to the table lookup model
  
• Develop different neural network models for
  different flight situations
• Use historical information to remove noise from
  probe readings

Discussion
What we need in next step
2
Apply reverse modeling to the table lookup model

• Preliminary results with reverse modeling
  algorithms do not indicate superior performance.
  
• This is because reverse modeling is highly
  dependent on the accuracy of the forward model.
• A neural network forward model is difficult to
  achieve optimal accuracy
• Using stochastic optimization techniques such as
  particle swarm or genetic algorithms is suggested

3
Develop different neural network models for
different flight situations
Probe Reading
Normal flight
Severe flight
Relationship between probes under different
volumes

• Probe readings behave different in severe flight
  situation
• Different flight situation have different
  specification accuracy
• Developing different neural network models for
  different flight situations may increase the
  precision of the network

4
Use historical information to remove noise
Probe Reading
Severe flight
Normal flight
Rules of probes reading vs. volume

• Probe reading should decrease continuously or
  keep a maximum or minimum value under same pitch
  and roll, caused by progressive losing of fuel
  during flight and ground
• Therefore, historical probe readings can be used
  to smooth current inputs or design filter to
  extract signal from noises.

5
Discussion

• Since the volume during flight or ground is a
  continuous parameter, using this powerful
  time-related information could reduce noise and
  increase the precise of the fuel measurement
• Improve the precision of the model
• Improve the neutral network itself, optimize the
  network
• Enhance the preprocessing of the raw data
• Add if-then rules according to experts
  knowledge
• Denoise raw data before input, which
  should be time-related

6
What we need in next step

• The table lookup model to build reverse model on
• Data under different flight or ground condition,
  distinguishing normal flight and relaxed flight
  or ground from severe flight
• Time-related probe readings to analyze the
  historical information we can use
• Practical data to study the feature of the noise
  in the field