• Citer

Contribution to the Efficient Monte Carlo Estimation of Curtailment Indicators in Active Network Management

par Sun, Juan
Président du jury Kinnaert, Michel
Promoteur Labeau, Pierre-Etienne
Publication Non publié, 2024-10-24

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

In the ongoing energy transition, the amount of decentralized production connected to the Medium Voltage (MV) power system is increasing. When the power produced is not consumed locally, reverse power flows will be injected into the High-Voltage (HV) grid. More frequent line congestions and voltage problems are thus likely to occur. To overcome this issue, an Active Network Management (ANM) scheme can be envisioned, whose aim is to control the injection of energy produced by Distributed Generation (DG) units to the grid, in almost real time, by possibly curtailing their production in case of grid congestion. This solution is as non-firm (or interruptible) connections offered to DG customers. In such a way, the power grid is always retained secure under ANM scheme, while facilitate more DGs integration into the current grid without network reinforcement. From the grid operator’s point of view, they prefer to get a quick evaluation on the maximum capability of DG units connection in the given node under ANM, while not affect too much the economic benefit of DG unit owners. In the previous deterministic “fit-and-forget” approach, assuming that the generators are able to output full capacity, the capability of MV buses to accept power injection is largely determined by the local minimum load and the export circuit capacity during the contingency of an outage of the highest-rated distribution circuit (i.e., classical N-1 criterion). This means that much of the distribution network exists without monitoring and control, and has spare capacity during normal operation. In contrast, flexible (or non-firm) connections, combined with a set of rules about the order that these generators are dispatched or curtailed under an ANM scheme, i.e., Principle of Access (PoA) rules, could increase the connection capability of distributed units in a power system. A typical grid loop is considered, where there are q DG units e.g., Wind Farms (WFs) and CHP (Combined Heat and Power) units connected to n substations. Once a new DG unit incorporation, there will be q+1 DG units. The impact of ANM scheme is assessed through a probabilistic methodology. Probability Density Functions (PDF/pdf’s) for their respective generation as well as for the loads at the different nodes of the power grid are first elicited. An efficient Monte Carlo sampling scheme, targeting electrically challenging scenarios which possible cause congestion events, is applied to generate variants (i.e., the samples of generations and loads), while using an Optimal Power Flow (OPF) to estimate the most economical curtailment for congestion case. The results are expressed with RI (i.e., Risk Indices or curtailment indicators), related with power curtailment and the probability of congestion occurrence, to expose the congestion risk. Meanwhile, the Utilization Factor (UF) of each unit displays the decrease against the more classical Capacity Factor (CF) after the corresponding curtailment management. This proposed algorithm only provides a performance assessment for an existing scenario (or a given grid). In order to estimate the evolution of the indicators (i.e., RI, UF of each unit) caused by the addition of one specific DG unit at a given node, it would be necessary to independently compute estimations of the indicators for both q case (i.e., initial grid) and q+1 scenario (i.e., grid after new DG unit connection). Furthermore, in case of a fictitious DG unit to be connected into a given node with a variable installed power, whose value is set from high-capacity to low-capacity until zero, the connection capacity will be derived from the multiple estimations of the same type of computations. However, reassessing the performance function for each scenario of data modeling will lead to a significant increase in computing load. And worse, as the difference between those indicators might be small, the results of the estimation will result in inaccuracy. Hence, this research further develops this proposed probabilistic assessment algorithm, in order to optimize the connected capacity in a MV grid, once selecting the most relevant nodes to which new DG units should be connected. Resorting to a Correlated Sampling (CS) method, as all variants (of loads and generations) with q DG units are physically possible with q+1 DG units, one computation process will simultaneously obtain the risk estimations for these cases with an acceptable accuracy. In like manner, the optimal connected capacity will eventually deduced from these candidate new fictitious DG units with the same type but different installed power. This will avoid repeating the timeconsuming estimation process and quickly provide an answer for possible connection request. Moreover, considering the variability of photovoltaic (PV) generation, whose probabilistic model is constructed by Kernel smoothing approach, improve and perfect this probabilistic algorithm.