This study is centred around the economic dispatch (ED) problem of thermal generating units over a multi-time interval, to have generators collectively meet the load demand and network losses while minimising the system fuel costs. The underlying optimisation problem displays non-convexity due to the consideration of the valve point effect and ramping rates limit of generators. Additionally, the system operator by itself cannot determine the behaviours of generators since they are not willing to share their own cost functions and wish to make decisions by themselves. In view of these reasons, the authors provide a decentralised method based on gradient projection and piecewise approximation for the ED problem. It is shown that the proposed method converges to the global optimum when applying appropriate step-size parameters. The performance of the proposed method is demonstrated on the IEEE standard 5 and 10 units test systems, and via comparisons with the existed methods in the literature.