Products, supplies, services, etc., of the power enterprise—an essential energy industry—are intrinsically linked to the production, operation, and long-term growth of every business, government agency, agricultural sector, service sector, and more. Clean energy, dependable energy, high-quality services, and energy conservation and emission reduction have all been elevated in recent years due to the ongoing power business reform and the development of industry regulatory rules. The internal control of power firms has been subject to increased pressures due to all of these factors. In this setting, it is crucial to analyze and mine huge data for vital indications of power firms’ internal control systems. This article presents the architecture of a machine learning-based neural network that can assess the quality of power firms’ internal controls. An enhancement to the DenseNet-based network is introduced in this article, which streamlines the network's structure and parameters while simultaneously enhancing the model's performance and the rate of convergence. Thirdly, to increase feature extraction's discriminative capabilities, the network implements an enhanced attention strategy. In the end, this essay uses the suggested network to assess the quality of power firms’ internal controls. This study presents experimental evidence that the suggested strategy for assessing the quality of power firms’ internal controls is effective.