Research And Application Of Adaptive Cuckoo Search Algorithm

Optimization and prediction problems play an important role in human society,and they are of great significance to power dispatching,traffic command and so on.Cuckoo Search(CS)is a bionic algorithm with the advantages of simple structure,easy deployment and so on.It is mainly used to solve combinatorial optimization problems.However,there exist low randomness of cuckoo population distribution,interference of out-of-range nests(invalid solutions that are not in the search range),complicated calculations and large fluctuations of Lévy flight,which are the main factors cause slow convergence.To solve these problems,this paper proposes an adaptive CS algorithm with historical experience and dynamic steps,and an adaptive CS algorithm with hyperchaotic system and expansion-mutation populations,which we call CHSACS and CEPCS respectively.Then,CEPCS is applied to optimize the echo state network(ESN),and the composition model of them is constructed and successfully applied to electric power load forecasting.The specific works are as follows.(1)Aiming at the slow convergence in CS due to the fixed parameters and the clumsy Lévy flight,the improved algorithm CHSACS is proposed.By introducing control parameters that change with iterations to adjust the Lévy flight step size and the update frequency of population,the algorithm can search within a larger solution space in the early iterations to improve the convergence speed,while in the later iterations it can search in a small range to improve the search accuracy.Because the out-of-range nests interference the convergence,we introduce a memory strategy to relocate the out-of-range nests in the search space to improve the stability of the algorithm.Experiments show that compared with original CS,PSO and ACO algorithms,CHSACS has faster convergence speed,higher search accuracy,and stronger ability to avoid local optima when dealing with continuous function optimization problems.(2)Aiming at the problems of slow convergence speed and low optimization precision caused by the complex calculation of Lévy flight,the fixed cuckoo population size and update frequency and low randomness of cuckoo population distribution,this paper proposes CEPCS algorithm.By constructing a new hyperchaotic system and generating random time series to replace Lévy flight,the algorithm structure is simplified and the convergence speeds up.By introducing expansion-mutation mechanism,CEPCS can adaptively adjust the distribution range and update frequency of the cuckoo populations according to the convergence situation,which effectively avoids from falling into the local optimum and improving the search accuracy.Through function testing and comparison with original CS and ACO algorithms,CEPCS has faster convergence speed,higher search accuracy and stronger ability to avoid local optima when dealing with complex multi-dimensional function optimization problems.(3)Aiming at the problems that the parameter setting of the ESN model relies on manual experience,which leads to poor versatility,slow training speed and unstable prediction accuracy,this paper proposes a combined model CEPCS-ESN using CEPCS to optimize ESN.First,we use the above CEPCS algorithm as the pre-layer,and optimize the parameters such as the weight coefficients of the ESN model as the optimization objective.Second,we apply the optimal parameter combination to construct the ESN model.And then the optimized ESN model is used for power load forecasting.Experiments show that compared with original ESN,ARIMA and EEMD-SVM models,CEPCS-ESN has faster training speed,better prediction accuracy,and less computation.