Two-Stage Day-Ahead and Intraday Active-Power Regulation of a Flexible Distribution Network Node: Scenario-Based Stochastic Optimization and Model Predictive Control

Distribution network operators increasingly require demand-side nodes to follow time-varying active-power instructions so that congestion and voltage problems can be mitigated locally. Industrial and commercial users are challenging participants in such schemes: they host substantial controllable electric equipment, yet their internal energy conversion processes must never be interrupted. This paper treats such a user as a flexible node of the distribution network, with the active power exchanged at the point of common coupling as the sole control interface, and proposes a two-stage day-ahead/intraday regulation framework, intended as a transferable scheduling model that parks with comparable resource portfolios can reuse by substituting their own load, renewable, device and price data. On the grid side, the user presents a portfolio of controllable electric resources: a gas-fired combined heat and power (CHP) unit acting as dispatchable local generation, electricity-driven power-to-heat and power-to-cooling units acting as adjustable electric loads, and storage that decouples the timing of the electric adjustment from the rigid internal demand, while the heat and cold balances are handled exclusively as hard internal constraints. The day-ahead layer is a scenario-based stochastic program built from Monte Carlo-sampled, K-means-reduced PV and wind scenarios; the intraday layer is a model predictive controller with a 15 min control interval and a 3 h receding window that corrects the deviation between the day-ahead expected renewable trajectory and the realized renewable output, together with device execution deviations modeled as zero-mean Gaussian perturbations. On a winter day benchmark, activating the full electric portfolio reduces the average node tracking error from 253.5 kW to the numerical zero level and raises the attainment rate from 13.5% to 100%, with the internal heat and cold balances holding as hard constraints on the set-point basis; on the realized basis, where the device execution disturbances are applied, the mean error stays between 12.4 and 13.6 kW and the heat imbalance is zero under the nominal disturbance, the strongest adversarial disturbance examined leaving 8.9 kW in three of the 96 intervals, i.e., 4.2 kWh or 0.05% of the daily heat demand. Under a time-of-use tariff with a 3:1 peak-to-valley ratio, the same portfolio produces a valley-heating, peak-generating schedule that cuts the daily grid purchase from 5.61 MWh to 0.33 MWh, and coordinating the battery on top lowers the daily operating cost by a further 4.7%.