Due to the intermittency of renewable energy generation, base stations (BSs) powered by renewable energy harvesters will experience power surpluses and power deficits over time. To use the renewable power more efficiently, energy harvesters with anti-correlated energy generation profiles should be deployed at BSs that are connected by transmission lines and power should be transmitted from surplus BSs to deficit BSs via transmission lines. In this paper, we develop an optimization algorithm to determine how energy harvesters with anti-correlated energy generation profiles should be deployed to every BS by taking into account the topology of the cellular network, i.e., whether or not a transmission line exists between a pair of BSs. Therefore, a BS is deployed more likely with an energy harvester type that is anti-correlated to those deployed at its connected neighboring BSs with our proposed algorithm. In addition, our proposed algorithm takes into account the distance-dependent power loss in the transmission lines. As a result, the shorter the transmission line between a pair of BSs, the more likely that these two BSs are deployed with anti-correlated energy harvesters. The renewable power that can be transmitted from the surplus BSs to the deficit BSs in the cellular network is on average around 40% higher with our proposed optimization algorithm in comparison with randomly deploying anti-correlated energy harvesters to the BSs.