Application Of Back Propagation Algorithm In The Coal Spectra Data Analysis

With the further development of the power industry of our country, about over 3/4 of theelectricity is accommodated by thermal power, the fuel that the newly-built large-scalethermal power plant used mainly bases on the coal, the quality of coal is directly related to theextent of the polluted atmosphere and the economic benefits of thermal power plant, etc..Therefore, through accurately analyzing the coal, we can judge the coal's nature, kind, anddifferent kinds of coal's processing effect and industrial use, make an appraise to the nature ofcoal. Consequently, we can reduce the air pollution degree and make greater economicbenefits. There are lots of traditional industrial analysis methods, especially with the NeutronInduced Prompt Gamma-ray Analysis (abbreviated as NIPGA) technology applied in the coalindustry, the technology of modern parsing spectra got great development too, but there existsome problems during the analysis course, for example, the non-linear problem andinterfering among the peaks. With the development of artificial intelligence technology,artificial neural network theory has got fast development too, Back Propagation neuralnetwork is one of the most representative neural network model among the various kinds ofneural networks, and has found extensive application. This paper analyzed the γ energyspectra of the coal by the function approach of BP neural network, and predicted its content ofcomposition. This paper expatiated the principle, characteristic and advantage of parsing spectrum byBP neural network, and the cause of not overcoming systematic error during analyzing theenergy spectra with this method. Meanwhile, it introduced BP neural network and BPalgorithm detailedly, analyzed the main defects and the reasons produced by this algorithm.Contraposing those difficult points and the defects of algorithm, this paper also put forwardthe corresponding improving measure, for instance: momentum item, adjusting the learningrate and momentum coefficient dynamically, and the method of "the succession studying",etc., designed the network structure and algorithm of parsing spectrum. While training the BPnetwork with the coal energy spectra, it adopted two activate function schemes: the activationfunctions of hidden layer and output layer are same or different, and look for the best hiddennode number of the training network respectively. Finally, it constructed the test networkstructures of these two schemes, and compared the content of composition which theirpredicted with the chemical method, then contrasted the precision of them. It was proved thatapplying the BP neural network in the coal energy spectra was an effective prediction method,and its precision was higher than other methods.