This work contributes to the research on three-phase photovoltaic systems via a new more robust online adaptive neuro-fuzzy maximum power point tracking (MPPT) control technique considering real partial shading and load conditions. The trapping in local minima and high computational cost in the existing neuro-fuzzy structure are addressed by incorporating B-spline function in the proposed control method. The system parameters are adjusted online via an adaptive neuro-fuzzy inference system rules acquired from the MPPT error. The optimization part of the proposed control law is performed through an online learning gradient-descent back-propagation algorithm. The superiority of the proposed control in terms of energy conversion efficiency, MPPT error, and output power is checked under the same operating conditions with well-known used traditional and intelligent MPPT control algorithms. Finally, the robustness of the proposed control is confirmed through a complete day simulation and comparison indexes.