The invention discloses a distribution type method for adjusting small base station transmitting power bias values in a self-adaptive mode. Under the condition that the return link capacity of a small base station is limited, the Q-learning method in machine learning is used for adjusting small base station transmitting power bias in the self-adaptive mode to share macro honeycomb loads, and the utilization rate of small base station return link resources is improved. Through the Q-learning method in machine learning, each small base station can conveniently monitor the number of peripheral users and changes of user distribution in real time, reference signal power bias values of the small base station are timely adjusted to obtain the optimal bias value selection under the current condition, and therefore the resources of the small base station are fully utilized for achieving the load sharing function for a macro bass station. The method can effectively improve system capacity and the covering effect, facilitates operation cost saving of the macro base station, and achieves green communication.