Abstract:
Coal is the cornerstone for energy security. In the current background of accelerating the development of the digital economy and actively and steadily promoting the “dual carbon” goal, the coal industry urgently needs to deepen digital transformation and intelligent construction. In this background, exploring the introduction of large model technology to empower coal industry applications, making full use of the industry’s massive knowledge data, and accelerating the digital development of the coal industry has become the focus of industry attention. Based on this, this paper sorts out the development status of generative large model technology, expounds the application status and effectiveness of large model technology in multiple fields, introduces the key technologies of industry large model such as data processing (cleaning, balancing, enhancement, etc.), text tokenization, pre-training and fine-tuning, prompt word optimization, vector embedding, alignment, retrieval enhancement generation and other large model technologies, and demonstrates that the industry large model inherits the advantages of the general large model of “general” and at the same time has the characteristics of “specialization”. This paper deeply analyzes the challenges of high R&D investment cost, difficulty in collecting high-quality data, and high difficulty in multimodal data fusion technology in the application of large model technology in the coal industry, and summarizes in detail the construction path and phased results achieved by SolStone Mine Large Model to cope with the above challenges from six aspects: infrastructure layer, data resource layer, algorithm model layer, application service layer, security and trustworthiness and testing layer, and industry ecological layer, and finally looks forward to the production and technological changes brought by the development of large model technology to the coal industry. It is pointed out that the construction of large models in the mining industry should follow the path of combining open access models and industry data, give full play to the tool attributes of large models to the application in scenarios, and build an application ecology combining “production-learning-research-application”, so as to help the development of new quality productivity in the mining industry.