Abstract as per original application
(English/Chinese): |
Optimal power flow (OPF) is a fundamental optimization problem, which aims to minimize the operational cost of a power network subject to its physical laws, resource capacities, and safety limits. The physical laws mainly refer to the nonlinear alternating-current (AC) power flow equations, which make the OPF problems nonconvex and thus hard to solve, especially with a large and ever-increasing number of decision variables for the rapidly growing controllable distributed energy resources.
In this project, we will develop new algorithms to solve large-scale nonlinear OPF problems faster than existing methods, with guaranteed feasibility and improved optimality of solutions. To realize this goal, the proposed algorithms will perform learning and computing in a spatially hierarchical and distributed structure on power distribution networks.
We will firstly develop a hierarchical distributed primal-dual gradient algorithm to accelerate the solution of large-scale linearized OPF problems. Then we will adapt this algorithm with an improved gradient calculation method to compensate for the model linearization errors and thus solve large-scale nonlinear OPF problems more accurately. To further enhance the speed of solving large-scale nonlinear OPF problems while ensuring the quality of solutions, we will learn decentralized neural network models from simulated data to predict the iterative descent directions of decision variables in the hierarchical distributed algorithm.
Each of the tasks above will start with a simplified, single-phase, radial (tree) power distribution network model and extend to the more realistic unbalanced three-phase model, with the latter containing power sources, loads, and transformers in mixed wye and delta configurations. We shall verify the solution quality and computational efficiency of the proposed algorithms through software simulation on power distribution network models with up to 100,000 nodes.
最優電力潮流 (OPF) 是一類基本的優化問題，旨在於物理規律、資源容量和安全限制約束之下，最小化電網運行成本。這裡的物理規律主要指非線性交流潮流方程，它們使得 OPF 問題具有非凸性，因而難以求解，尤其是在可控分佈式能源快速增長且決策變量個數不斷增加的情況下。在本項目中，我們將開發新的算法，以比現有方法更快的速度求解大規模非線性 OPF 問題，同時保證解的可行性並提高其最優性。為實現這一目標，所提出的算法將立足於配電網的空間分層分佈式結構進行學習和計算。我們將首先開發一種分層分佈式原-對偶梯度算法，以加速大規模線性化 OPF 問題的求解。然後，我們將採用改進的梯度計算方法對此算法進行調整，以補償模型線性化誤差，從而更準確地求解大規模非線性 OPF 問題。為進一步提高大規模非線性OPF問題的求解速度並確保解的質量，我們將從仿真數據中學習分散式神經網絡模型，以預測分層分佈式算法中決策變量的迭代下降方向。上述每個任務都將從簡化的單相徑向（樹形）配電網模型開始，並擴展到更真實的三相不平衡模型，後者包含混合星形和三角形連接的電源、負載和變壓器。我們將通過對多達 100,000 節點的配電網模型進行軟件仿真，驗證所提算法的求解質量和計算效率。
|
