The real world is full of problems with multiple conflicting objectives. However, reinforcement learning traditionally deals with only a single learning objective. Recently, several Multi-Objective Reinforcement Learning (MORL) algorithms have been proposed. Nevertheless, many of these algorithms rely on tabular representations of the value function which are only suitable for solving small-scale problems. In addition, although there are some existing MORL techniques can learn the Pareto optimal solutions, they can only be applied to the simple multi-step problems in the discrete environment. However, many real-world problems are involving in the continuous and stochastic environment. Therefore, an effective MORL technique is needed to address these problems. Evolutionary Computation algorithms are such an effective method for multi-objective problems and support different solution representation such as the rule-based solution and the neural network-based solution. Moreover, the rule-based solution has good scalability and it is easy to understand, whereas the neural network-based solution is more compact and effective for handling the continuous inputs. Therefore, this research aims to learn the Pareto optimal policies for the deterministic, continuous and stochastic MORL problems with the rule-based and the neural network-based representations.