As global energy demands and climate concerns continue to grow, the need for renewable energy is becoming increasingly clear and wave energy conversion (WEC) systems are receiving growing interest. Over the years, many WEC systems have emerged but most of these systems are designed to extract energy from a single direction of motion. In reality, there are six degrees of freedom that a conversion device can potentially harness and a device that can operate on multiple axes, should be able to more effectively and consistently produce power. However, the design and control of the power take off (PTO) system for a multi-axis device is challenging due to the system complexity and nonlinearity. WEC systems often utilize optimal control techniques for PTO operation and leverage a prediction of the upcoming wave force to ensure power optimization. Prior work has clearly demonstrated that high power production can be achieved when an exact system model is used and the upcoming wave conditions are known, but uncertainty in the underlying model or the wave prediction can degrade performance. PTO control on a multi-axis WEC must leverage predictions of forces in multiple directions and if model predictive strategies are used, must leverage a simplified model of the WEC dynamics to be able to optimize in real time. The uncertainty in these predictions and the model could severely degrade the WEC’s power output. This work examines the control of a multi-axis WEC system, TALOS, and leverages machine learning to predict wave forces over the upcoming time horizon. TALOS is a point-absorber type WEC with multi-axis PTO system. The PTO uses a heavy ball that is attached to the hull with springs and hydraulic cylinders. When the hull is pushed by the external waves, the relative motion between the ball and hull moves the hydraulic cylinders causing them to pump a fluid through a circuit, thereby driving a hydraulic motor to produce electricity. This design has shown promising results in energy output but is more challenging to control since the PTO can move over six degrees of freedom. This paper seeks to quantify wave prediction uncertainty and its seasonal variation and to examine the impact of the uncertainty of the prediction on a model predictive controller’s ability to optimize the power output of TALOS.