An Online Reactive Power-Optimization Strategy Based on Load-Curve Prediction and Segmentation

Abstract

Due to fluctuating characteristics of loads, dynamic reactive power optimization over a certain time period is essential to provide effective strategies to maintain the security and economic operation of distribution systems. In operation, reactive power compensation devices cannot be adjusted too frequently due to their lifetime constraints. Thus, in this paper, an online reactive power optimization strategy based on the segmentation of multiple predicted load curves is proposed to address this issue, aiming to minimize network losses and at the same time to minimize reactive power-compensation device adjustment times. Based on forecasted time series of loads, the strategy first segments each load curve into several sections by means of thresholding a filtered signal, and then optimizes reactive power dispatch based on average load in each section. Through case studies using a modified IEEE 34-bus system and field measurement of loads, the merits of the proposed strategy is verified in terms of both optimization performance and computational efficiency compared with state-of-the-art methods. optimization is of importance, which optimizes for a time period and makes the adjustment times of devices one of the optimization objectives.Dynamic reactive power optimization is a large-scale multi-period mix-integer non-linear programming problem. The classical methods such as non-linear programming [4] require differentiable objective function and are thus not applicable for dynamic optimization.A number of attempts have been made to conquer the problem of dynamic optimization. Reference [13] made daily reactive power optimization considering network losses, adjustment times of on-load tap changer (OLTC) and switchable capacitor banks, using an improved multi-population ant colony algorithm. Reference [14] used mixed integer non-linear programming to optimize transmission losses only, considering constraints of reactive power equipment operation limits and power grid security. Using a filter collaborative state transition algorithm (FCSTA), reference [15] built a two-objective dynamic reactive power optimization model, with one objective to minimize the actual power losses and voltage deviations, the other to minimize incremental system loss compared with the last time stamp. The objectives can be exchanged during dynamic optimization.Generally, state-of-the-art attempts include reducing scales of objectives or constraints, optimizing for increments at continuous time stamp, and using AI methods. However, much computational burden will be introduced when using AI methodology.Aiming to maintain objectives and constraints, as well as to reduce computational complexity, this paper proposes to segment load curves in a duty cycle into several stable time series first, and then conduct optimization for each segment. Based on online load predictions, the method can be applied in an online mode. Since the whole time series of loads need to be segmented into parts with lowest inner deviation, the methods of wind ramp extraction ...