A novel grey model is developed for electricity consumption forecasting. Relationship of initial value and time input is investigated. The time response function is improved by a heuristic optimization. The new approach adapts to the time-varying characteristic of raw data. The proposed model outperforms the results of other existing models.