Abstract: Battery energy storage system (BESS) is one of the most effective resources to compensate demand-supply imbalance induced by renewable energy and frequency security. With large-scale access to the power grid, BESSs coordinate with thermal power units to participate in peak regulation and frequency regulation. To this end, BESSs need to reserve sufficient capacity in advance. The required frequency regulation capacity is determined by the implicit relationship between frequency variations and power imbalance. Existing methods linearize the frequency regulation capacity model according to the mixed integer linear programming characteristics of the unit commitment problem. Such linearization methods guarantee the computational efficiency at the cost of an increased approximation errors. In view of the trade-off between model accuracy and computational efficiency, this paper develops a model between frequency and power based on the Universal Approximation Theorem of deep neural network (DNN), which do not change the problem structure of unit commitment and might approach the original model with an arbitrary accuracy in theory. Based on such methods, we propose a joint scheduling method for battery energy storage in unit commitment. First, the mechanism of multi-source joint frequency regulation is analyzed, and the control strategy of energy storage to provide fast power support frequency modulation in response to load step disturbance is proposed. According to the control strategy, a DNN is constructed to predict that the system can withstand the maximum disturbance, and the standby method of frequency modulation capacity of energy storage is proposed. Second, the joint scheduling model of battery energy storage peak regulation and frequency regulation is constructed to optimize the unit operating state and the capacity allocation of energy storage peak regulation and frequency regulation. Last but not least, case studies based on IEEE 39-bus system are conducted to verify the proposed work. The results show that the proposed method can ensure the frequency security of the system under the maximum step disturbance, and improve the economy of the system operation, which proves the superiority of the data-driven method in the expression of frequency fluctuations.

Key words: battery energy storage, deep neural network (DNN), peak regulation, frequency regulation, unit commitment