A large number access of wind and photovoltaic power generations will cause fundamental changes in the planning and operation characteristics of distribution networks. This paper proposes a capacity selection method based on board learning system. The board learning capacity configuration model is trained by using data such as network node voltage and power output. The accuracy of the model and the rationality of the results are evaluated by using root mean square error and voltage stability evaluation indicators. The IEEE33 node system is used as an example to simulate. The capacity selection results, which satisfy the total investment cost and the network active loss, are obtained. And compare with the support vector machine and the kernel extreme learning machine to verify the feasibility and effectiveness of the proposed model and method.