Machine learning based energy use prediction
Techniques for using machine learning to predict the energy use of mobile actors such as a moving person or a mobile machine, as well as technologies for using such predictions. Energy use prediction for a moving person or a mobile machine can include an energy use prediction for a person or machine moving from one place to another regardless of the complexity of the corresponding route and the complexity of other variables involved. Mobile machines can include any type of mobile machine for any industrial or consumer use. Mobile machines can include watercraft, aircraft, terrestrial vehicles, automobiles, or robots, for example.
1 . A method, comprising:
generating, by a computing system, a speed-time profile for a route from a starting point to a destination point, based on a first model and a task profile used as an input for the first model;
predicting, by the computing system, an energy-use profile for the route, based on a second model and the speed-time profile used as an input for the second model;
using, by the computing system, the energy-use profile to control a hybrid vehicle according to data in the energy-use profile, wherein the hybrid vehicle comprises two energy sources, and wherein the control of the hybrid vehicle comprises controlling, by a controller of a hybrid vehicle, use of the two energy sources during the route according to the data in the energy-use profile;
repeating the generation of the speed-time profile and predicting of the energy-use profile in a statistical analysis to generate a statistical distribution; and
using the statistical distribution of the energy-use profile to further control the hybrid vehicle according to data in the statistical distribution, and wherein the control of the hybrid vehicle further comprises controlling, by the controller of the hybrid vehicle, use of the two energy sources during the route according to the statistical distribution of the energy-use profile.
2 . The method of claim 1 , further comprising:
using the statistical distribution of the energy-use profile to generate a graphical user interface according to data in the statistical distribution;
emulating a state of the hybrid vehicle controlled according to the energy-use profile;
determining a prediction error in the prediction of the energy-use profile based on a difference between the emulation of the state of the hybrid vehicle and a predicted state of the hybrid vehicle in the energy-use profile;
updating the first model or the second model according to the prediction error;
reusing the first model and the second model after the update to enhance the accuracy of the energy-use profile; and
providing the prediction error in the graphical user interface graphically.
3 . The method of claim 2 , further comprising using the statistical distribution of the energy-use profile to generate a distribution of a probability of completion corresponding to the statistical distribution of the energy-use profile.
4 . The method of claim 3 , further comprising using the distribution of the probability of completion to control the hybrid vehicle and to generate the graphical user interface according to data in the distribution of the probability of completion.
5 . The method of claim 4 ,
wherein the generating the speed-time profile is also based on an actor profile used as a second input for the first model, and
wherein the actor profile and the task profile are two separate inputs of the first model.
6 . The method of claim 5 , wherein the actor profile comprises a hybrid vehicle profile input.
7 . The method of claim 6 , wherein the predicting the energy-use profile comprises the second model predicting energy consumption over the route for each chronological step of the speed-time profile.
8 . The method of claim 6 ,
wherein the predicted the energy-use profile is also based on the task profile used as a second input for the second model, and
wherein the speed-time profile and the task profile are two separate inputs of the second model.
9 . The method of claim 6 ,
wherein the predicting the energy-use profile comprises using a road-load determination, and
wherein the road-load determination is enhanced through machine learning to capture terrain-specific energy use prior to its use in the predicting of the energy-use profile.
10 . The method of claim 6 ,
wherein the first model generates a route-optimized drive cycle, and
wherein the speed-time profile is generated according to the route-optimized drive cycle.
11 . The method of claim 6 ,
wherein the repeating of the generation of the speed-time profile and the prediction of the energy-use profile is a part of a Monte Carlo simulation, and
wherein the method further comprises using the Monte Carlo simulation to generate the statistical distribution of the energy-use profile.
12 . A method, comprising:
generating, by a computing system, a speed-time profile for a route from a starting point to a destination point, based on a first model and a task profile used as an input for the first model;
predicting, by the computing system, an energy-use profile for the route, based on a second model and the speed-time profile used as an input for the second model;
using, by the computing system, the energy-use profile to control a mobile machine according to data in the energy-use profile, wherein the control of the mobile machine comprises controlling, by a controller of the mobile machine, use of an energy source during the route according to the data in the energy-use profile;
repeating the generation of the speed-time profile and predicting of the energy-use profile in a statistical analysis to generate a statistical distribution;
using the statistical distribution of the energy-use profile to further control the mobile machine according to data in the statistical distribution, and wherein the control of the mobile machine further comprises controlling, by the controller of the mobile machine, use of the energy source during the route according to the statistical distribution of the energy-use profile;
using the statistical distribution of the energy-use profile to generate a graphical user interface according to data in the statistical distribution;
emulating a state of the mobile machine controlled according to the energy-use profile;
determining a prediction error in the prediction of the energy-use profile based on a difference between the emulation of the state of the mobile machine and a predicted state of the mobile machine in the energy-use profile;
updating the first model or the second model according to the prediction error;
reusing the first model and the second model after the update to enhance the accuracy of the energy-use profile; and
providing the prediction error in the graphical user interface graphically.
13 . The method of claim 12 , wherein the predicting the energy-use profile comprises the second model predicting energy consumption over the route for each chronological step of the speed-time profile.
14 . The method of claim 13 ,
wherein the predicted the energy-use profile is also based on the task profile used as a second input for the second model, and
wherein the speed-time profile and the task profile are two separate inputs of the second model.
15 . The method of claim 14 ,
wherein the predicting the energy-use profile comprises using a road-load determination, and
wherein the road-load determination is enhanced through machine learning to capture terrain-specific energy use prior to its use in the predicting of the energy-use profile.
16 . A method, comprising:
generating, by a computing system, a speed-time profile for a route from a starting point to a destination point, based on a first model and a task profile used as an input for the first model,
wherein the generating the speed-time profile is also based on an actor profile used as a second input for the first model,
wherein the actor profile and the task profile are two separate inputs of the first model, and
wherein the actor profile comprises a mobile-machine profile input;
predicting, by the computing system, an energy-use profile for the route, based on a second model and the speed-time profile used as an input for the second model;
using, by the computing system, the energy-use profile to control a mobile machine according to data in the energy-use profile, wherein the control of the mobile machine comprises controlling, by a controller of the mobile machine, use of an energy source during the route according to the data in the energy-use profile;
repeating the generation of the speed-time profile and predicting of the energy-use profile in a statistical analysis to generate a statistical distribution; and
using the statistical distribution of the energy-use profile to further control the mobile machine according to data in the statistical distribution, and wherein the control of the mobile machine further comprises controlling, by the controller of the mobile machine, use of the energy source during the route according to the statistical distribution of the energy-use profile.
17 . The method of claim 16 , wherein the predicting the energy-use profile comprises the second model predicting energy consumption over the route for each chronological step of the speed-time profile.
18 . The method of claim 17 ,
wherein the predicted the energy-use profile is also based on the task profile used as a second input for the second model, and
wherein the speed-time profile and the task profile are two separate inputs of the second model.
19 . The method of claim 18 ,
wherein the predicting the energy-use profile comprises using a road-load determination, and
wherein the road-load determination is enhanced through machine learning to capture terrain-specific energy use prior to its use in the predicting of the energy-use profile.
20 . The method of claim 18 ,
wherein the first model generates a route-optimized drive cycle, and
wherein the speed-time profile is generated according to the route-optimized drive cycle.