Performance of Photovoltaic Maximum Power Point Tracking Algorithms in the Presence of Noise Alexander M. Latham, Charles R. Sullivan, and Kofi M. Odame Thayer School of Engineering Dartmouth College Hanover, NH 03755 Email: alexander.m.latham@dartmouth.edu, charles.r.sullivan@dartmouth.edu Abstract—This paper introduces a probabilistic analysis of the effects of noise on various maximum power point tracking (MPPT) algorithms for photovoltaic systems, including how noise affects both tracking speed and overall efficiency. The results of this analysis are verified by simulations. This analysis provides a better understanding of how noise affects performance and it can be used to optimize an MPPT system. I. Introduction Maximum power point tracking (MPPT) has become a standard technique for high-performance photovoltaic sys- tems. An intelligent controller adjusts the voltage, current, or impedance seen by a solar array until the operating point that provides maximum power for the connected array in the present temperature and insolation conditions is found. There is a large body of literature describing MPPT control techniques, including the surveys [1] and [2]. Although the established techniques are routinely implemented in industry, and generally give satisfactory performance, pub- lication on the topic continues to accelerate, with dozens of publications per year in the last decade [1], in part because of the importance of getting the best possible output from an expensive solar array. Key metrics for an MPPT algorithm include tracking speed and accuracy, as is discussed extensively in the literature (e.g., [3]). However, the fundamental constraint on tracking accuracy is often the effect of noise in the measurement on the behavior of the tracking algorithm. Noise can also affect tracking speed in some cases. Standard tracking algorithms involve directly or indirectly intro- ducing a periodic perturbation in the operating point in order to measure the slope of some characteristic. This perturbation reduces the power obtained from the solar panel because the panel is no longer operated consistently at the maximum power point, even if the algorithm has successfully found that point. This provides an incentive to reduce the size of the perturbation. However, as the size of the perturbation is reduced, the signal-to-noise ratio in the measurement of the slope is degraded. Thus, noise fundamentally limits the performance. This is particularly important in methods that require a current measurement, as some current measurement methods (e.g. Hall-effect transducers) are inherently noisy [4], and the use of a sense resistor entails a tradeoff between signal-to-noise ratio in the measurement and power loss in the resistor [5]. The importance of noise is acknowledged in a subset of the literature on MPPT and is sometimes used to motivate particular algorithms or hardware configurations (e.g., [3], [6]–[10]) but with very few exceptions (e.g., [11]), the impact of the noise is not analyzed quantitatively. In [12], we quantitatively analyze the effect of noise on a continuous-time MPPT algorithm. In this paper, we develop quantitative analysis of the impact of noise on two discrete-time maximum power point tracking algorithms. The analysis is verified through dynamic simulations which include noise, and the performance of these algorithms are compared to the system analyzed in [12]. II. Noise Effect on Slew Rate of Perturb and Observe Consider an MPPT system with a simple perturb and observe (P&O) tracking algorithm, where one changes a variable X, which could be a voltage, current or duty cycle, that influences the operating point of the array, by a fixed ΔX each period, ΔT , and measures the power output of the array to determine how to change X next [1], [2]. The slew rate, how fast the algorithm will move toward the MPP, will be influenced by the amount of noise in the measurement of power. The maximum slew rate for the algorithm is ΔX ΔT . However, with the addition of noise to the system, wrong decisions may sometimes be made about whether to increase or decrease X, leading to a slower average slew rate. For this analysis, the noise considered is Gaussian white noise that shows up on the power measurement of the array. We assume the signal representing the output power is integrated during the period between decisions, and so the standard deviation of the noise being added to each measurement of power is σ n = k/ √ ΔT , where k is a constant with units volts/ √ Hz. When the system makes a decision about whether to increase or decrease X, it looks at the change in power from the previous step to the current step (ΔP). At each point on the power vs. X curve, the signal that will be seen is mΔX, where m is the slope of the curve. In order for the algorithm to make the wrong decision about whether to increase or decrease X, the noise must have a magnitude greater than mΔX and a sign opposite to that of the slope. Also, as the signal used, mΔX, comes from two measurements, the standard 978-1-4244-5287-3/10/$26.00 ©2010 IEEE 632