Combining random forests and physics-based models to forecast the electricity generated by ocean waves: A case study of the Mutriku wave farm

This paper combines random forests with physics-based models to forecast the electricity output of the Mutriku wave farm on the Bay of Biscay. The period analysed was 2014–2016, and the forecast horizon was 24 h in 4-h steps. The Random Forest (RF) machine-learning technique was used, with three sets of inputs: i) the electricity generated at Mutriku, ii) the wave energy flux (WEF) prediction made by the ECMWF wave model at Mutriku's nearest gridpoint, and iii) ocean and atmospheric data for the Bay of Biscay. For this last input, extended empirical orthogonal functions (EOFs) were calculated to reduce the dimensionality of these data, while retaining most of the information. The forecasts are evaluated using the R-Squared, Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The model easily outperforms a persistence forecast at 8–10 h and beyond. The most accurate forecasts are achieved by using all three of these inputs. This approach may help to effectively integrate wave farms into the electricity market. •The electric power generated at the Mutriku wave farm was predicted for a 24-h forecast lead time.•The RF model proved to be a better prediction tool than Persistence itself beyond 8–10 h.•This is the first time the output power of a fully operational wave farm has been forecast up to a 24-h ahead horizon through the use of logged electric power data.•Wave energy needs to maximise the accuracy of the forecasting methods to solve the intermittency and variability problems associated with renewable energies and the study presented here makes an initial contribution to this objective.