•
A data-driven scheduling-correction framework is proposed to provide real-time actions with a global vision, making better use of the seasonal hydrogen storage. In the scheduling module, ex-post optimal SoC sequences of the seasonal storage are generated using historical data. In the correction module, the real-time operation relies on a bi-objective rolling-horizon optimization problem which minimizes the operating cost while following the SoC reference of the seasonal storage, which is sequentially updated along with time.
•The kernel regression technique is established to sequentially update the SoC reference of the seasonal hydrogen storage. This technique calibrates a conditional probability distribution based on the newly observed data; the resulting distribution (weight coefficients) interprets how likely the current year is similar to each of the past years. The reference is updated in each period according to the weighted sum of the ex-post optimal SoC sequences.
•Systematic tests are conducted to compare the cost, load shedding, and renewable energy curtailment achieved by different methods. Numerical results show that the scheduling-correction framework can fully exploit the seasonal storage. The proposed method outperforms existing rolling horizon approaches in compromising economy, power supply reliability, and renewable energy utilization.