Predicting fuel burn and emissions for aircraft
A method for fueling an aircraft comprises (a) predicting a sequence of fuel-burn quantities over a planned flight based on at least one sequence of multivariate flight data recorded during a prior flight and on a hidden state of a trained machine. For each of a pre-selected series of prior flights, a corresponding sequence of multivariate flight data recorded during the prior flight is processed in a trainable machine to develop the hidden state, which minimizes an overall residual for replicating the fuel-burn quantities in each corresponding sequence. The method further comprises (b) summing the fuel-burn quantities over the planned flight to obtain a fueling estimate; and (c) fueling the aircraft based in part on the fueling estimate.
1 . A method for fueling an aircraft, the method comprising:
predicting a sequence of fuel-burn quantities over a planned flight based on at least one sequence of multivariate flight data recorded during a prior flight and on a hidden state of a trained machine,
wherein for each of a pre-selected series of prior flights, a corresponding sequence of multivariate flight data recorded during the prior flight is processed in a trainable machine to develop the hidden state, which minimizes an overall residual for replicating the fuel-burn quantities in each corresponding sequence;
summing the fuel-burn quantities over the planned flight to obtain a fueling estimate; and
fueling the aircraft based in part on the fueling estimate.
2 . The method of claim 1 wherein, for each of the pre-selected series of prior flights, the corresponding sequence spans a climb phase, a cruise phase, and a descent phase of the prior flight.
3 . The method of claim 2 further comprising, for each of the pre-selected series of prior flights, adjusting the corresponding sequence to provide a common climb length for the climb phase, a common cruise length for the cruise phase, and/or a common descent length for the descent phase.
4 . The method of claim 3 wherein adjusting each corresponding sequence comprises padding the corresponding sequence in the climb phase, the cruise phase, and/or the descent phase.
5 . The method of claim 1 ,
wherein replicating the fuel-burn quantities comprises transforming the multivariate flight data from a previous time step in each corresponding sequence based on the hidden state, and
wherein predicting the sequence of fuel-burn quantities comprises transforming the multivariate flight data from a previous time step in the at least one sequence based on the hidden state.
6 . The method of claim 1 wherein at least the trained machine comprises a trained encoder arranged logically upstream of a trained decoder, wherein the encoder is trained to emit a vector that featurizes an input sequence of multivariate flight data, and wherein the decoder is trained to replicate the fuel-burn quantities of the input sequence based on the vector, thereby generating an output sequence.
7 . The method of claim 6 wherein the encoder and the decoder are configured according to a long short-term memory (LSTM) architecture.
8 . The method of claim 7 wherein at least the trained machine comprises an encoder-decoder LSTM (E-D LSTM) model, a convolutional neural-network (CNN) LSTM encoder-decoder (CNN LSTM E-D) model, or a convolutional LSTM encoder-decoder (ConvLSTM E-D) model.
9 . The method of claim 6 wherein at least the trained machine further comprises a fully connected layer configured to interpret the fuel burn quantities at each time step of the output sequence.
10 . The method of claim 1 wherein the at least one sequence and each corresponding sequence of multivariate flight data includes flight-data recorder (FDR) data selected via principle-component analysis.
11 . The method of claim 1 wherein the at least one sequence and each corresponding sequence of multivariate flight data includes atmospheric data.
12 . The method of claim 1 wherein the at least one sequence of multivariate flight data share a departure airport, an arrival airport, and an aircraft specifier of the planned flight.
13 . The method of claim 1 wherein the trainable machine, after development of the hidden state, is the trained machine.
14 . The method of claim 1 wherein the trainable machine is distinct from the trained machine, the method further comprising:
receiving, in the trained machine, a plurality of parameter values defining the hidden state as developed on the trainable machine.
15 . The method of claim 1 wherein the planned flight is one of a plurality of candidate flights of a plurality of candidate routes between a departure airport and an arrival airport, the method further comprising:
for each of the plurality of candidate flights:
predicting a sequence of fuel-burn quantities over the candidate flight based on at least one sequence of multivariate flight data recorded during a prior flight and on the hidden state of the trained machine, and
summing the fuel-burn quantities over the candidate flight to obtain a flight-wise fueling estimate;
for each of the plurality of candidate routes, summing every flight-wise fueling estimate along the candidate route to obtain a route-wise fueling estimate for the candidate route,
wherein the planned flight is a flight on the candidate route with a lowest route-wise fueling estimate.
16 . A trained machine comprising:
an input engine configured to receive at least one sequence of multivariate flight data recorded during a prior flight;
a prediction engine configured to predict a sequence of fuel-burn quantities over a planned flight based on at least one sequence of multivariate flight data recorded during a prior flight and on a hidden state of the trained machine,
wherein for each of a pre-selected series of prior flights, a corresponding sequence of multivariate flight data recorded during the prior flight is processed in a trainable machine to develop the hidden state, which minimizes an overall residual for replicating the fuel-burn quantities in each corresponding sequence
a summation engine configured to sum the fuel-burn quantities over the planned flight, to obtain a fueling estimate; and
an output engine configured to output the fueling estimate.
17 . The machine of claim 16 wherein the output engine is further configured to output an emissions estimate based on the fueling estimate.
18 . A trainable machine comprising:
an input engine configured to receive, for each of a pre-selected series of prior flights, a corresponding sequence of multivariate flight data recorded during the prior flight;
a training engine configured to process each corresponding sequence to develop a hidden state, which minimizes an overall residual for replicating the fuel-burn quantities in each corresponding sequence;
an output engine configured to expose at least a portion of the hidden state.
19 . The machine of claim 18 wherein each corresponding sequence includes flight-data recorder (FDR) data selected via principle-component analysis.
20 . The machine of claim 18 wherein for each of the pre-selected series of prior flights, the corresponding sequence spans a climb phase, a cruise phase, and a descent phase of the prior flight, the trainable machine further comprising:
an adjustment engine configured to adjust each corresponding sequence to provide a common climb length for the climb phase, a common cruise length for the cruise phase, and a common descent length for the descent phase.