Uncertainty-Aware Drone Dispatch: Recording-Energy Prediction and Charging-Constrained Simulation

Abstract How much service does a drone depot give up when it allows for error in a mission’s energy forecast? We analysed 181 of 209 recordings from a DJI Matrice 100 flown on a common route in Pennsylvania. A ridge regression predicted whole-recording electrical energy from planned speed, payload and altitude. We estimated a one-sided allowance from the largest underprediction on each of 4 calibration dates. Across 34 recordings on 4 reserved dates, mean absolute error was 0.650 Wh and root mean squared error was 0.755 Wh . Every recording on each date fell below its nominal 80% upper estimate. A separate discrete-event simulation compared dispatch rules based on point and upper energy estimates from the development data. In 864 paired replications across 18 hypothetical depot scenarios, the upper rule completed fewer requests in every scenario and reduced reserve breaches in 16 . For the same 2,259 requests in the reference scenario, upper estimates reduced completions from 977 to 955 and reserve breaches from 5 to 0 . An open-access workbench lets readers vary the available charge and inspect the resulting dispatch decision.