A brain-inspired spiking neural network for the renewable energy short circuit ratio prediction

With the skyrocket of renewable energy integration, the scale of long-distance power transmission and consumption has gradually expanded. The voltage support strength of the generation system is easily affected by the unexpected fluctuation of the onsite renewable energy stations, which seriously affects the power system stability. Previously, the short-circuit ratio (SCR) is widely utilized for grid strength detection, as it can reflect the situation clearly and intuitively. Therefore, the renewable energy short-circuit ratio (RESCR) is constructed to keep the step with the power system transformation. The RESCR is a crucial indicator for power shifting and protection scheme design. However, due to the undesired fluctuation of renewable power, the RESCR deviates more severely than the conventional SCR index. Thus, the RESCR prediction research is significant for regulating and operating the power generation system. The RESCR prediction methods mostly use traditional artificial neural networks with large power consumption. To solve this problem, we proposed a prediction model based on residual structure and spiking neural networks called SpikeNet, which has the characteristics of event-driven and sparse representation and can realize low-power computing. Based on this model, we also proposed a novel training framework that includes a pre-training process and an online prediction process. In the pre-training process, the network can learn basic prediction knowledge based on historical data. The network can obtain continual learning capability through secondary learning in the online prediction process. Based on the utility-side historical data, we construct a dataset used for pre-training and then verify the proposed prediction model's online function.