Construction and Optimization of Deep Learning-based Innovative Experimental Platform for Mathematical Intelligence in Electricity Major

Abstract

Due to the lack of management capability of the model development process on traditional experiment platforms, they cannot meet the continuous experimentation, reproducibility, and traceability needs of university researchers in the model development process. For this reason, the authors propose research on the design and optimization of a platform for digital intelligence innovation experimentation that is combined with deep learning. Through a theoretical analysis of the key technologies used in the deep learning development platform, the authors summarise the design structure of the deep learning innovation experiment platform. The algorithm management module, training management module, and model deployment module primarily construct this structure. Aiming at the problem of the slow computation speed of the platform in large data sets and high-dimensional space, a gradient descent algorithm is used to optimise the platform. The platform is validated and analyzed in terms of performance testing and application effects. The results show that the experimental platform optimized by the gradient descent algorithm has a higher throughput than the traditional experimental platform, and the difference is 0.1~1.5. Also, class A’s total score on experimental reflection ability (2.58±0.877) was significantly higher than class B’s (1.48±0.377) after the experiment, and the p-value was less than 0.05, which showed that the new experimental platform was more likely to improve students’ experimental reflection ability than the old way of teaching.