Framework and Function Design of Machine Learning Platform for Power Generation Process

With the transformation of China’s energy structure, the control requirements for the power generation process are becoming increasingly precise and intelligent. Applying machine learning algorithms to improve operational efficiency has become one of the important demands of current power generation enterprises. However, the deployment of various machine learning technologies that have already been put into operation in the power generation process is independent of each other, and the data interaction ability with the control system is limited. This not only makes the station lack of a unified operations and maintenance management entrance, but also requires a large amount of manual operation to complete the entire process from model training to inference, making it difficult to update the model. This paper proposes a framework and function design of the machine learning platform for the power generation process aming at the data characteristics and control system operation mode of the power generation production site. It provides a programming and management interface based on algorithm modules in a graphical interface. At the same time, the platform’s ability to obtain historical and real-time data of the production site is achieved by coupling the training program and inference program with the control system, so that all machine learning processes can be included in the platform’s automated operation scope. The application of this platform can facilitate users to complete customized model training for selected production processes, efficiently complete machine learning tasks without affecting the normal operation of the control system, reduce the manpower and learning costs of applying machine learning technology, and provide a feasible solution for power generation enterprises to improve their intelligent operation level.