Abstract: Operational effectiveness is a key indicator to measure the effectiveness of weapons. It is a simple and effective way to improve the operational effectiveness of weapons by finding operational effectiveness sensitive indicators. In order to solve the problems of high computation cost and low computation velocity of complex evaluation model， this paper introduces the extreme learning machine as the agent model to replace the complex performance evaluation model. To improve the operational effectiveness of weapons， this paper uses variance-based global sensitivity analysis to find the key factors that affect the effectiveness of weapons， and then find the equipment associated with its function to be improved. This paper uses the typical combat mission of the submarine as the effectiveness optimization case to evaluate the feasibility of the method. Compared with the global sensitivity analysis based on the feed-forward neural network agent model and the support vector regression model, the experimental results verify the validity and efficiency of L-EML model.

Key words: global sensitivity analysis, extreme learning machine, surrogate model, performance optimization