Abstract:

An optimized control strategy was proposed for Solid Oxide Fuel Cells (SOFC) with two-layer compound considering temperature-gradient transient-response to online operate SOFC efficiently and safely. In the optimization layer, a neural network was used to predict the power change with a particle swarm optimization (PSO) algorithm being used to obtain the optimal operation curve of the system under different powers. In the control layer, considering the transient fluctuation of load, a compound controller was used to track the optimal operating point of the system. The results show that the root mean square error of the short-term and long-term memory (LSTM) neural network is reduced by 20.2 W, and the prediction accuracy is higher than that of the nonlinear autoregressive neural network (NARNN); The compound controller and the traditional proportion-integration-differentiation (PID) controller have good control effects on the stack temperature, the fuel utilization rate, and the air over-oxygen ratio; When the load power changes rapidly, the temperature gradient transient response of the compound controller is smaller, and it is below the threshold to avoid excessive temperature gradients in the system.

Key words: solid oxide fuel cell (SOFC), online optimal control, transient response, compound controller, nonlinear auto regressive neural network (NARNN), long short-term memory (LSTM) neural network