Renewable energy prediction: A novel short-term prediction model of photovoltaic output power

Highlights

• A hybrid improved MVO (HIMVO) intelligent optimization algorithm is proposed.

• The HIMVO is used to optimize the parameters of support vector machine (SVM).

• The HIMVO-SVM model has a better convergence speed and higher prediction accuracy.

Abstract

Photovoltaic power generation is gradually developing into a massive power industry with the maturity of renewable energy power generation technologies. Photovoltaic power generation is greatly affected by external factors and the output power is characterized by randomness and indirectness, which poses a great challenge to photovoltaic grid-connection. A hybrid improved multi-verse optimizer algorithm (HIMVO) is proposed to optimize the support vector machine for photovoltaic output prediction. HIMVO algorithm introduces chaotic sequences to initialize the population, which significantly enhances the convergence rate of the algorithm compared with the multi-universe optimizer algorithm. This study applied particle swarm optimization algorithm, dragonfly algorithm, multi-universe optimizer algorithm and HIMVO to testify the availability of the hybrid improved multi-verse optimizer support vector machine model (HIMVO-SVM). The results indicate that HIMVO algorithm has better optimization ability and stability. The four models, HIMVO-SVM, multi-verse optimizer support vector machine, particle swarm optimization support vector machine, back propagation and radical basis function neural network are used to predict output in three different weather types. The results indicate that the model has higher prediction accuracy and stability. The mean square error value of the HIMVO-SVM model decreases by at least 0.0026, 0.0030 and 0.0012, and the mean absolute percentage error value decreases by at least 3.6768%, 1.9772% and 2.7165%, respectively. The proposed method is beneficial to the prediction of output power and conduces to the economic dispatch of the grid and the maintenance of the stability of the power system.

Introduction

Energy and environmental issues have become the common concern of all over the world. With the gradual exhaustion of fossil energy and the increasing severity of environmental pollution, making full use of renewable energy will be a necessary way to solve energy and environmental problems (Lee et al., 2017). Compared with fossil energy, renewable energy generally has the characteristics of small pollution and large reserves. The use of renewable energy can effectively alleviate the consumption of fossil energy, reduce environmental pollution, and is conducive to the healthy development of social economy (Tseng, 2017). As a kind of renewable energy, solar energy has attracted more and more attention due to its rich energy storage, wide distribution, green and clean features, and has been recognized as the best alternative energy solution globally (Mandal et al., 2012). Photovoltaic (PV) power generation is an effective way to use solar energy, but also an important part of new energy generation, with broad application prospects.

With the continuous improvement of PV power generation technology, the scale of PV power grid connection is expanding day by day. However, PV power generation has the characteristics of randomness, indirectness and volatility affected by climatic factors. (ELAMIM et al., 2018, Tascikaraoglu et al., 2016). These features bring many problems to the PV grid-connection, and also impact the operation and scheduling of power grid (Ehsan et al., 2017). Zhang et al. (2015) presented the accurate output prediction has become an important work, still, in different forecasting time scales, short-term forecasting, for instance, hours to days have an important decisive effect on real-time grid dispatching and has a direct impact on the safety and stability of system operation. Hu et al. (2018b) proposed a three-stage prediction model to illustrate the impact of ultra-short-term PV power prediction. Therefore, this study analyzes the short-term prediction of PV output power.

There have been some previous studies on the prediction of PV output power. Wang et al. (2017) adopted two prediction models of support vector machine (SVM) and K-Nearest Neighbors for PV output power prediction according to the size of training sample data. Liu et al. (2018) used genetic algorithm to optimize three different types of neural networks, and combined the PV output according to different weights. Esen et al. (2017) used adaptive neuro-fuzzy inference system and artificial neural network models to predict the performance of solar ground source heat pump system. Zang et al. (2018) proposed a prediction method based on convolutional neural network, which can mitigate the influence of solar radiation uncertainty on prediction effect. Hu et al. (2018) used an indirect prediction method. The dynamic characteristics of the cloud in the sky were extracted, the input data was processed according to different meteorological features, and then the PV output was predicted by the radical basis function (RBF) neural network.

Aforementioned, those methods are applied to the prediction of PV output power. However, PV power generation is greatly affected by weather types. Prior studies only analyze a single weather type and do not consider different weather conditions. Indirect prediction methods cannot learn directly from historical data, and prediction models are complicated. In addition, the prediction performance of models such as propagation (BP) and extreme learning machine is greatly affected by the selection of their own parameters. At the same time, the demand for training data is large, and the performance when the number of data is small is not satisfactory. Therefore, in order to simplify the prediction process and further improve the short-term prediction accuracy of PV output power, the SVM model with few self-parameters and strong generalization ability for small samples and nonlinear problems is selected as the prediction model. In addition, the multi-verse optimizer (MVO) intelligent optimization algorithm is selected and improved to optimize SVM parameters, and a hybrid improved multi-verse optimizer support vector machine (HIMVO-SVM) prediction model is established. The HIMVO-SVM model is applied to the PV output power prediction experiment under different weather types, and the experimental results are compared with other models to prove the validity of the proposed model. Hence, the contributions are as follows.

• The prediction model is put forward to predict PV output power and the test result indicates the model meets the precision requirement;

• The prediction model possesses better convergence speed and accuracy compared with other prediction models;

• This study to improve the capacity of PV consumption and increasing the utilization of solar energy resources;

• This study is conducive to the power generation department to arrange power generation plans, optimize grid dispatching, and rationally arrange grid operation modes;

• This study is beneficial to improve the quality of PV grid connection and reduces the impact of PV output volatility on power system operation.

The structure of this study is organized as follows. Chapter 2 provides and discusses the literature review of PV output prediction. Chapter 3 describes the methods used and establishes the prediction model. Chapter 4 discusses the influencing factors of PV power generation. Chapter 5 introduces the experiment and analyzes the experimental results in detail. Chapter 6 discusses and analyzes the experimental results of prediction. Chapter 7 introduces the conclusion and the contribution.