In conventional evolutionary algorithms, different search strategies have been designed to find the optimum in a fitness landscape. Nevertheless none of them works well over all possible fitness landscapes. Since the fitness landscape associated to a complex global optimization problem usually consists of various local landscapes， each search strategy is efficient in some particular type of fitness landscapes. Hence, a novel approach is proposed in the project, called mixed strategy self-feedback evolutionary algorithms based on local fitness landscapes. It may dynamically adjust search strategies according to local landscapes and choose a search strategy best fitting a local landscape with a probability from a strategy pool. In order to achieve such adaptation, machine learning techniques, including reinforcement learning and supervised learning, will be applied to the learning of the best mixed strategy for a specific local landscape. Mixed strategy evolution algorithms will be applied to the problems of soil composition detection, water-saving irrigation system optimization design, modelling, estimation of crop optimal irrigation, the soil water characteristic parameters of soil nutrient management and fertilization model partition optimization. An embedded evolutionary system will be designed in order to implement real-time detection, monitoring and control in precision agriculture.