The typical characteristic values of electrical signals in plant from time domain, frequency domain and time-frequency domain were analyzed and the electrical signals in plant were to be as physiological indicators. To establish environment prediction models, typical features of electrical signals and some environmental parameters were chosen to be as input of neural network with the extreme learning machine algorithm characterized by fast learning speed and good generalization. The results showed that the plants electrical signals were the low-frequency weak signals by analysis of the electrical signals in Peperomia tetraphylla leaf on different domains, and by extreme learning machine three prediction models such as temperature, humidity and illumination were established to make plants grow well. Compared with the traditional BP neural network, the root mean square error with ELM algorithm is less than 0.97, while the coefficient of determination is more than 0.92 and each training time is less than 0.03s. This method provided the scientific basis for greenhouse environmental regulating and was verified to be feasible.