Prediction of overflow concentration of thickener based on ISSA-LSTM

The monitoring of the overflow concentration of the thickener is the key to realize intelligent dosing of coal slurry. The overflow concentration monitoring method based on the sensor will lead to the delay of flocculant regulation. In order to solve the above problem, a prediction method of overflow concentration of thickener based on improved sparrow search algorithm (ISSA) and long-short term memory (LSTM) is proposed. Firstly, the correlation analysis and pretreatment of multi-parameter time series in the process of concentration production are carried out to obtain the input variables. Secondly, the multi-strategies are combined to improve sparrow search algorithm (SSA). Tent chaotic map is introduced to initialize the sparrow population to ensure population diversity and speed up algorithm convergence. The optimization process of SSA is improved by using the spiral predation strategy to balance both local development and global search capabilities. The firefly perturbation strategy is used to perturb the sparrow search results to improve the global search performance and avoid the algorithm falling into local optimization. Thirdly, ISSA is used to optimize the hyperparameters of the two-layer LSTM network model. Finally, the overflow concentration prediction model based on ISSA-LSTM is established for on-line monitoring. The experimental results show the following points. ① The Ackley function and Rastigin function are selected as test functions. It is concluded that ISSA's global optimization capability and convergence speed are better than those of the particle swarm optimization (PSO) algorithm, whale optimization algorithm (WOA) and standard SSA. ② Among the three improved strategies, the spiral predation strategy plays a leading role in improving the performance of ISSA. The chaotic map and the firefly perturbation strategy coordinate the convergence speed and global search capability of the algorithm to further improve the optimization performance of the algorithm. ③ ISSA is used to optimize the hyperparameters of LSTM, which solves the problem of under-fitting or over-fitting when the values are determined by subjective experience. The prediction precision of overflow concentration of the ISSA-LSTM model reaches 97.26%, which is higher than that of double-layer LSTM, SSA-LSTM, and least square support vector machine (LSSVM) models. ④ Data pretreatment can improve the precision of the model, and the prediction precision of overflow concentration after noise reduction is improved by 30.25% compared with that before noise reduction.