Research On Knowledge Acquisition For Colliery Gas Forecast Based On Rough Sets And Neural Networks

Gas explosion is the most common disasters in coal mine production, and the primary governing means of gas explosion are the gas forecast at present. Expert systems have many the advantages and features to meet the needs of gas forecast, for example strong adaptability and reliability, quick responsibility, persistent, composite and explanation etc. However, knowledge acquisition is the key and bottleneck question in building colliery gas forecast expert system, which have greatly restricted the development and application of colliery gas forecast expert system. In order to solve this problem, this thesis studies knowledge acquisition for colliery gas forecast.According to the advantages and the problems exiting in rough sets and neural networks, this thesis presents a knowledge acquisition model for colliery gas forecast, which consists of six modules: data preprocessing, neural network construction, neural network training, rules extraction, rules verification and rules input. The overall model structure and main function module are presented in detail; the knowledge acquisition algorithm is investigated. It is applied in the real-time data through the simulation experiment, and results shows that it solves the problem of the knowledge acquisition difficulty of colliery gas forecast, and has good real time character, high reliability and perfectly precision. The model provides good foundation for the establishment of knowledge base of colliery gas forecast.The main works in this paper includes: 1. With analyzing and investigating rough sets and neural networks, a knowledge acquisition model for colliery gas forecast is proposed, which combines two kinds of intelligence technology.2. A knowledge acquisition algorithm is presented. Firstly, the algorithm uses rough sets to select the data for a neural networks structure and learning. Secondly, the hidden layer unites of the neural networks is modified dynamically in the process of neural networks training, the redundant and unimportant hidden unites and links are deleted from a trained neural networks, ultimately get a better neural networks. Finally, the rules are extracts by an algorithm from neural networks.3. The model is realized by Visual Basic and Matlab, and the ability of knowledge acquisition is testified with the real-time data.