Abstract: This paper proposes a decision method using least square support vector regression（LS-SVR）to find the optimal situation.Algorithm models have been developed to predict the output power of PV plants with dust accumulation,as well as the peak sun hours according to weather classification. The economic benefits of PV plants could be evaluated dynamically for different time schedules using these models,and thus the best cleaning time during specific period is just the one that maximizes the benefits. Field experiments are carried out to verify the feasibility of our method. When the results evaluated from measured data are compared with those evaluated from prediction data,it is clear that our method is capable of predicting the optimal cleaning time of PV plants. Besides,it is also noticed that the economic benefits of optimized cleaning is also related to the length of period and cost of cleaning. The method proposed in this work provides a promising tool for the real-time optimization of dust cleaning strategy,which would be of great meaning during the maintenance and operation of PV plants.