Adaptive estimation of optimal combination weights for partial-diffusion Kalman filtering together with its mean convergence and stability analysis is proposed here. The simulations confirm its superior performance compared with the existing combiners. Sensor networks with limited accessible power highly benefit from this design.