An Improved Sparrow Search Algorithm Based on Complementary Inertia Weight and LSTM Optimization for Timer Serial Forecasting

When dealing with optimization problems that require to achieve optimal result within fewer iterations, the basic Sparrow Search Algorithm (SSA) shows insufficient solution accuracy and poor stability. In view of the above shortcomings, an improved adaptive Sparrow Search Algorithm based on complementary inertia weight (CIW-SSA) is proposed. Firstly, considering the lack of population diversity, chaos mapping is used to generate all individuals. Secondly, in order to obtain high-precision optimization results within short-term iterations, this paper created and implemented a pair of complementary inertia weights to improve search efficiency. The cooperation of this pair of complementary inertia weights not only adjusts the inertia of each individual, but also adjusts the step size of movement, which greatly improves the speed and accuracy of convergence. Thirdly, to response to the alert value of basic SSA and improve the capacity of exploration, Gaussian mutation and Cauchy mutation are organically added together to process of position update of producer. Simulation experiments were carried out through 18 classical benchmark functions with different characteristic, and the result show that the improved algorithm has faster convergence speed, more stable convergence properties and higher convergence accuracy. In addition, to prove the superiority of CIW-SSA within a few iterations, CIW-SSA is employed to optimize the hyperparameter of Long Short-Term Memory (LSTM) model for wind speed forecasting, and the test results prove the effectiveness of the improvement strategies.