Research And Implementation Of Multi-Objective Optimization Algorithms For Decision Support

Workflow task scheduling in a cloud computing environment is a common application scenario in the field of intelligent decision making,and in which the scenario requires the allocation and scheduling of several parallel computational tasks to distributed computing nodes.Since the scenario involves several objective dimensions that need to be optimized simultaneously,this thesis applies a multi-objective optimization algorithm to solve the problem.In the existing related research,there are still problems such as low efficiency of the optimization algorithm and easy to fall into local convergence,on the other hand,the existing work still has problems such as incomplete expression of preferences and high time consumption of algorithm interaction operation in the performance of the algorithm fitting the decision maker's preferences.To solve the related problems,this thesis focuses on the following three issues in the improvement of multi-objective evolutionary algorithm and interactive multi-objective optimization algorithm.First,an improved multi-objective evolutionary algorithm incorporating elite strategies is proposed,which uses meta-heuristic search to encode workflow scheduling schemes,maps the encoded schemes to the decision space,and solves the optimal workflow task scheduling scheme by iterative optimization with an improved evolutionary operator after incorporating elite strategies based on multiple groups.This algorithm enables to improve the quality of understanding while ensuring the convergence speed of the algorithm and avoiding falling into local convergence through innovative improvements of the process framework and evolutionary operator of the multi-objective evolutionary algorithm.The underlying algorithmic framework and the required model data support are provided for the subsequent research problems.Second,an interactive multi-objective optimization algorithm with preference embedding is proposed for the problem of expressing and embedding the subjective preference information of decision makers.The algorithm is based on the desired reference points set by the decision maker and an interactive preference vector generation method based on multiple constraints,which together constitute a comprehensive preference model of the decision maker and guide the evolutionary direction of the multiobjective optimization algorithm,so as to obtain the optimal solution set for optimization.Then,by improving the knee point detection method,the point with the least loss of compromise for each decision objective dimension is selected as the optimal solution output that meets the decision maker's preferences.Finally,the subsystem implementation of workflow scheduling tasks for a cloud computing platform is carried out.In this thesis,the system implementation is first analyzed for the overall flow and algorithm module design of the system.The system allows the decision maker to set the computation nodes and workflow task attributes,as well as the dependency constraints between subtask nodes,as the initial inputs to the multi-objective optimization algorithm.The decision maker can also select the desired reference points according to his own preferences,generate a decision maker preference model through an interactive algorithm,and finally obtain a scheduling solution that meets the decision maker's preferences.Aiming at the problems of low efficiency of algorithms and poor preference of decision makers in other existing work,this paper proposes two optimization algorithms,and provides sufficient support for the decision process of workflow task scheduling through the optimization of the algorithm.Compared with other existing algorithms,this paper can generate a better solution set in a shorter convergence time,and the final output solution can also better fit the decision maker's preference,resulting in a smaller decision error.