To solve the problems that the traditional fruit fly algorithms fall into convergence too early and the results are not stable

an improved fruit fly optimization algorithm based on global-local bidirectional driving is proposed. Firstly

in order to comprehensively consider the global driving information of fruit fly population and the local driving information of a fruit fly individual

the conceptions of advanced group and memory space are introduced. In each iteration

the fruit flies which have good performances are defined as the advanced group

and the historical best positions of a fruit fly are defined as the memory space of this fruit fly. Secondly

in order to avoid the premature convergence problem

the global driving effect of the fruit flies in the advanced group is considered

and the dimensional components of the fruit fly position vectors are updated sequentially in the position updating processes. Finally

in order to avoid the blind global searching when the population approaches convergence

each fruit fly will consider the local driving effect of its own cognitive experience

and the roulette strategy is used to select the positions in the memory space for jumping out the local optimum. The experimental results of typical test functions and the web anomaly detection simulation show that

the proposed fruit fly optimization algorithm based on global-local bidirectional driving has high searching accuracy

good stability

and high convergence speed

and is suitable for dealing with the complex problems with high dimensions in web anomaly detection.