Improving the accuracy of load identification is the key technology to realize non-intrusive load monitoring. Based on the problems of existing models, such as low recognition accuracy, high feature redundancy, and poor separability, this paper proposes a load recognition method based on random forest (RF) and genetic algorithm optimized extreme learning machine (GA-ELM). First, the time domain and frequency domain information of the steady-state current signal are extracted as load features. In order to further reduce the redundancy of the feature set and eliminate the poorly separable features, the RF algorithm is used to optimize the features and obtain the optimal Feature set. Finally, the genetic algorithm is used to optimize the weight and bias param-eters of the extreme learning machine, and the load identification model is established. The 16 load states of 11 household appliances are identified by the model. The experimental results show that the model proposed in this paper can increase the recognition accuracy, and the use of this model can identify household load quickly and effectively.