The total load forecasting of urban power network is the fundamental work of urban power network planning. In order to fully tap and use more information of load historical data, a bi-directional prediction method for the annual maximum of total load forecasting in urban network is proposed. The method analyzes of the correlation of power load and power consumption based on historical data, establishes the model of Load-electricity ratio, hereby obtains the power load maximum of each month by using power consumption data, and then forecasts the total load maximum of the objective years on the basis of these maximum from horizontal and vertical directions by using linear regression, exponential smoothing and gray theory, serves the weighted average value of the six predicted value as the Final predicted results. Example analysis shows that the method is correct and effective.