The search strategies were constructed with the combination of three main topics, which were sequential decision problem, approximate methods (or heuristics) and optimisation. These topics were defined as following:
Sequential decision problem: SDDPs are characterized as making multi-stage decisions sequentially over time. Successive Markov decision tree, which is called MDP, has been widely used to describe stochastic multi-stage decision problems whose transition mechanism can be controlled over time. Due to the dynamic, stochastic and combinatorial nature of the given problem, the problem can be classified as an NP-hard decision problem. Therefore, the search strategy was constructed to cover broadly not just for sequential decision problems but also any problem which can be depicted by an MDP or classified as any NP-hard decision problem.
Approximate methods/Heuristics: Heuristics aim to identify near-optimal solutions at a reasonable computational time by introducing certain pre-set subjective factors. As heuristics do not guarantee to find a global optimum, the term “heuristics” often means approximate methods in the context of optimisation. In this study, thus, “heuristics” are equivalent to approximate optimisation method. Heuristics also include meta-heuristics, which are a set of heuristic methods aimed at efficiently and effectively exploring a search space.
Optimisation: Optimisation is the process to find the solutions that give the maximum or minimum value of the objective function under given circumstances[46]. The process of solving SDDPs can be broadly viewed as optimisation, which is the act of obtaining the best result under the given circumstances.
