Recently, Northwest China has been suffered PM2.5 pollution. Through a case study of Lanzhou city, capital of Northwest China’s Gansu province, a hyperchaotic cuckoo search-extreme learning machine (HCCS-ELM) approach is proposed to establish a PM2.5 concentration prediction model. The cuckoo search is used to generate the number of hid- den layer neurons in ELM. The hyperchaotic system is innovatively introduced to improve the accuracy of the algorithm with its good ran- domness and convergence. HCCS-ELM has a similar memory footprint as ELM, while experiments using observed data from four monitoring stations in Lanzhou during 2016 to 2018 show that HCCS-ELM has higher accuracy in PM2.5 hourly and daily concentration prediction.