Optimal Allocation of Land Area for a Hybrid Solar Wind Power Plant Neha Sengupta, Kaushik Das, T.S. Jayram, Deva P. Seetharam IBM Research, India neha.sengupta@in.ibm.com, kausdas8@in.ibm.com, t.s.jayram@in.ibm.com, dseetharam@in.ibm.com, Abstract—Recently, there has been a growing interest in wind solar hybrid power plants as a means to overcome the inherent intermittency in both resources. One crucial decision faced by a hybrid plant designer is to determine an allocation of land area between wind and solar installations so as to satisfy the generation requirements in terms of the power output by the plant along with the availability. In this paper, a methodology is proposed for this optimal allocation problem under the constraints of climate and weather conditions, land area available for renewable energy harvesting, and PV/ wind turbine charac- teristics. Based on historical wind speed and solar insolation data for a given location, probabilistic models using parametric and non parametric estimation techniques are developed to capture the variability and periodicity in these resources. A probabilistic model for the hybrid system is also derived using which the optimal allocation that satisfies the generation requirements is determined. The proposed method is validated by performing a detailed experimental analysis on historical data for an arbitrary location on the globe. I. I NTRODUCTION Renewable energy resources, such as wind and solar, are considered highly promising in the face of growing concerns for the environment, energy conservation, and sustainable development [1]. Daily as well as seasonal variability are inherent to both wind and solar resources [2]. Traditionally, the uncertainty of a standalone solar panel or wind turbine installation is managed using a storage system. However, this results in an increased overall cost of the output energy, and therefore limits the benefits of using renewable energy [3]. A hybrid solar wind energy system uses two renewable energy sources, thereby improving the system efficiency and power reliability, and reducing the energy storage requirement [4]. However, aggregating inherently stochastic power sources such as wind and solar to achieve reliable electricity supply is a non trivial problem [5]. Consider the scenario of a typical market model described in [6], wherein each day at a specified hour, the owner of the plant establishes a schedule of the net power export to the grid for each hour of the next day. Power is traded in a day-ahead spot market, and any deviations from the contracted power due to forecast errors are settled in a balancing market. For the hybrid plant owner who bids in this market, this has serious implications in terms of managing the risk induced by the un- certain renewable energy sources. It follows that the question of characterizing the reliability of a given hybrid installation must be answered for meaningful market participation. This is also relevant to the planning phase of a hybrid plant, in which the sizing of wind and solar installations must be optimized in order to maximize magnitude and availability of energy output. The availability of wind and solar energy at a given location depends on climate and weather conditions and is highly variable [4]. Therefore, a prediction model for these resources should be stochastic in nature to be able to account for the inherent variability. A majority of the existing tools for forecasting these renewable resources give a deterministic forecast, also known as a spot or a point forecast. This is a single value for each forecast horizon. Such methods suffer from the drawback that they provide no information about any departure from the prediction. In decision-making applications based on stochastic optimization or risk assessment, a point forecast finds limited use. It has been shown that for trading future production on an electricity market as described above, using probabilistic predictions rather than point forecasts yeilds greater benefits [3]. In [7] First Order and Second Order Markov Chain Models are used to develop a purely statistical and probabilistic wind power forecasting method. Both models are applied on a wind power dataset and their performances are compared with that of a Persistent Model. [3] uses non-parametric estimation techniques to build probability density functions for wind speed forecasting. AI techniques like artificial neural networks, fuzzy logic, genetic algorithms, and wavelet neural networks have been used for solar power prediction. [8] gives an overview of all these methods as applied to the problem of PV sizing. Various optimization techniques for hybrid PV/wind sys- tems sizing have been proposed in the literature. Wang et. al. [5] discussed the applicability of the concepts of stochastic network calculus to analyze the achievable level of system reliability with appropriate number of PV cells, wind turbines, and energy storage capacity. One source of difficulty in ap- plying stochastic network calculus is obtaining reliable supply and demand models. Tina et al. [2] presented a probabilistic approach based on the convolution technique to assess the long-term performance of a hybrid solar-wind power system (HSWPS) for both stand-alone and grid-linked applications. They generated the probability distribution functions (pdf) for wind and solar power using parametric estimation methods. The size of PV and wind are chosen such that the energy that cannot be supplied to fulfill the demand is minimized. An optimization problem is formulated to choose the most cost- effective combination. Terra et. al. [9] proposed a procedure to obtain the optimal sizing of a grid connected HSWPS