A LSTM Prediction Method Optimized by Improved Sine and Cosine Algorithm

In view of the problem that the long and short term memory network is sensitive to the setting of hyper-parameter, and considering that the application object of this study is the coke and circulating air volume in the Coke Dry Quenching (CDQ) production process, which is easily affected by the complex and harsh production environment. In this paper, a LSTM prediction model based on improved sine and cosine algorithm was proposed. Based on the standard LSTM network, the Haar wavelet transform was used to improve the sine and cosine optimization algorithm, and then the improved sine and cosine algorithm was used to optimize the hyper-parameter of LSTM network, including the learning rate \( \alpha \), batch size and epoch size. The ISCOA-LSTM prediction model based on time series data was constructed. Finally, the coke and circulating air volume of three different CDQ furnaces were taken as research objects, and tehe coke and circulating air volume were predicted using the method proposed in this study. The experimental results show that, compared with the manual calculation method, the accuracy of the method proposed in this study is higher. The accuracy of the prediction of coke was increased by 7\( \mathrm{{\% }}\)–12\( \mathrm{{\% }}\), and the accuracy of the prediction of circulating air volume was increased by 7\( \mathrm{{\% }}\)–10\( \mathrm{{\% }}\). The prediction accuracy of this method is higher, which can overcome the influence of the complex environment in CDQ production, and provides an effective method for the prediction of coke and circulating air volume in CDQ production.