Optimized deep belief network based on an improved Blood-sucking Leech Optimizer algorithm for wastewater quality forecasting.

To address the issues of limited exploration capability and premature convergence in the optimization process of the Blood-Sucking Leech Optimizer (BSLO) algorithm, we propose an Improved BSLO (IBSLO) algorithm. Initially, a directional leeches switching mechanism based on an inverted S-shaped nonlinear perceived distance to strike a balance between exploitative and exploratory capabilities of the algorithm. Subsequently, a dynamic perception signal was designed to simulate dynamic stimulus signals, guiding leeches to search and optimize more accurately. Finally, the memory sharing mechanism is incorporated to improve search efficiency and secure the global optimal solution of the algorithm. In addition, the IBSLO algorithm is assessed through 23 benchmark functions and the standard test set from CEC-2017, with its superiority confirmed by a detailed analysis of the algorithm's convergence. To further assess the efficacy of the IBSLO algorithm in addressing practical optimization challenges, it was utilized to enhance the predictive model for crucial water quality parameters within the wastewater treatment procedure. The IBSLO-Deep Belief Network model's prediction results demonstrated superior accuracy compared with other optimization strategies, further confirming the excellent performance of the IBSLO algorithm.