In the coming years, ensuring the electricity supply will be one of the most important world challenges. Renewable energies, in particular wind energy, are an alternative to non-sustainable resources thanks to their almost unlimited supply. However, the chaotic nature and the variability of the wind represent a significant barrier to a large-scale development of this energy. Consequently, providing accurate wind power forecasts is a crucial challenge. This paper presents AMAWind, a multi-agent system dedicated to wind power forecasting based on a cooperative approach. Each agent corresponds to a turbine at a given hour, it starts from an initial production forecast and acts in a cooperative way with its neighbors to find an equilibrium on conflicting values. An assessment of this approach was carried out on data coming from a real wind farm.