摘要
•We present a dimensionality reduction approach for many-objective Markov Decision Processes.•A lower dimensional representation of the original objectives is identified via Non-negative Principal Component Analysis.•The approach is demonstrated on the operation of a multi-purpose water reservoir (Tono Dam, Japan) involving 10 objectives.•The NPCA-based solutions provide a good representation of the 10 objective Pareto front in terms of consistency and diversity.