Coordinating heterogeneous multi-electrolyser systems under renewable forecast uncertainty often requires repeated solution of large dispatch optimisation problems, which limits the online scalability of conventional stepwise rolling dispatch. However, a newly arrived control interval does not necessarily invalidate the multi-step dispatch plan generated by the previous optimisation solve. Event-triggered adaptive-horizon model predictive control (ET-AH-MPC) is developed as a selective re-optimisation framework that decides at each control step whether the existing multi-step dispatch plan should be reused or refreshed. A mixed-integer linear programming (MILP) dispatch model is formulated for heterogeneous alkaline and proton-exchange-membrane electrolysers with five-state transitions to optimise state transitions and power allocation. Event-triggered refresh evaluates stored-plan validity using forecast mismatch, operating-boundary activation, and plan exhaustion, while adaptive-horizon selection determines the look-ahead depth only after a refresh is triggered. Execution-layer recourse converts the selected forecast-based command into an executable command under realised wind power. In case studies with real wind-power data and closed-loop forecast mismatch, ET-AH-MPC retains 98.55% of net profit and 98.63% of hydrogen production relative to a stepwise model predictive control baseline. It also reduces optimisation refreshes and cumulative solving time by 77.86% and 85.85%, respectively. Selective re-optimisation allocates MILP computation to intervals where the stored plan shows validity risk, rather than to every control-execution interval.