Search

Large-scale integration of renewable generation has changed the way that distribution networks are operated, posing significant challenges to voltage control. To address this challenge, system operators need to exploit the potential of inverter-based renewable generation for reactive power regulation. However, the distribution network parameters are usually inaccurate or incomplete, making conventional centralized or model-dependent distributed control methods difficult for fast voltage tracking and optimal reactive power distribution. This paper proposes a model-free distributed Newton method for voltage control based on data-driven lift-dimension linear power flow which does not reply on accurate and complete network parameters. By using matrix splitting to calculate Hessian matrix, the proposed method possesses a super-linear convergence and exhibits superiority in fast response over existing linear methods. Furthermore, a Koopman-based state space mapping method is proposed to obtain global sensitivity and tune the iteration direction in an off-line manner, which can realize model-free voltage control. The convergence and optimality of the proposed method are validated by case studies. Especially, the parameter independence feature of the model-free scheme provides decided superiority in scenarios of incomplete model. Even under communication failures, the proposed method still can maintain voltage stability at a suboptimal point.

The high penetration of distributed generators (DGs) deteriorates the uncertainty of active distribution networks (ADNs). Soft open points (SOPs) can effectively improve flexibility and deal with operational issues in ADNs. However, the formulation of SOP control strategies depends on the accurate mechanism model. Data-driven method can utilize only measuring data to conduct operation and becomes a promising way. In practical conditions, the measuring data may suffer from bad data and measuring errors, which poses a challenge to meet the diverse operational requirements. This paper proposes a data-driven multi-mode adaptive control method for SOP with measuring bad data. First, considering the inaccurate network parameters and quality of measuring data, a robust data-driven framework for SOP operation is proposed based on robust hierarchical-optimization recursive least squares (HO-RLS). Then, a multi-mode control strategy for SOP is proposed to adapt to the diverse operational requirements. A dynamic triggering mechanism is designed to achieve adaptive mode switching. The case studies on practical distribution networks show that the proposed method can fully explore the benefits of SOP to improve the operational performance of ADNs. The potential limitations are discussed to enhance practicality.

The proliferation of distribution-level phasor measurement units (D-PMUs) with a high reporting rate brings a heavy transmission burden to communication systems of distribution networks, which necessitate efficient data compression on edge computing devices. This paper proposes a real-time D-PMU data compression algorithm, including three stages of prediction, quantization, and Bitpack. The current data frame is predicted by the adaptive normalized least mean square predictor based on the stochastic gradient descent algorithm. Then, the prediction errors are quantized as integers and Bitpack is established to extract significant bits and losslessly reduce the redundancy of the quantized data. For edge computing devices accessing multiple D-PMUs in distribution networks, a performance optimization mechanism is proposed. The spatial similarity of D-PMUs is explored to multiplex the predictor and reduce the computation burden. In addition, the compression performances can be adaptively adjusted to diminish the transmission delay in limited bandwidth conditions. Finally, the proposed method is validated and compared with the state-of-the-art methods using the field data and simulated data in normal and fault conditions of distribution networks. Moreover, the impacts and countermeasures of data quality are considered. The results demonstrate that the proposed method achieves accurate and efficient real-time compression in different scenarios.

Data-driven power flow (PF) analysis methods have higher practical value to deal with the inaccurate branch parameters in medium and low-voltage distribution networks, which can be easily integrated into the classic optimization framework as a constraint. However, the PF mapping relationship of the measurement data shows stronger nonlinearity, with the high penetration of distributed generation (DG), leading to the low accuracy of the linear PF constraint. To this end, we propose an incomplete dimension lifting (IDL) data-driven PF method for distribution network optimization. The state variables are divided into optimization variables and uncontrollable variables, among them, the optimization variables keep the original linear state space or converted into quadratic space to simplify the optimization solution; and the uncontrollable variables is dimension lifted to fit the nonlinearity of PF. In addition, in order to deal with the problem of gradient direction error caused by nonlinear model overfitting, the first-order sensitivity constraint is involved in parameter regression process. The standard IEEE cases with high penetration of DG demonstrate that the proposed method can achieve higher accurate PF constraint and better optimization results, compared to the existing data-driven PF constrain.