China
Africa
Explore chapters and articles related to this topic
Electricity Pricing
Published in Mohammad Shahidehpour, Muwaffaq Alomoush, Restructured Electrical Power Systems, 2017
Mohammad Shahidehpour, Muwaffaq Alomoush
In this chapter, we have reviewed some basic concepts in electricity price forecasting, such as price calculation and price volatility. Because of its importance, we also discussed the issue on factors impacting electricity price forecasting, including time factors, load factors, historical price factor, etc. We used the artificial neural network method to study the relationship between these factors and price. We proposed a more reasonable definition on MAPE to avoid the demerits of traditional methods on measuring forecasting in the context of electricity price forecasting. Practical data study showed that a good data pre-processing was helpful, i.e., using too many training vectors or considering too many factors is not good for price forecasting. Practical data study also showed adaptive forecasting could improve forecasting accuracy. We concluded that the artificial neural network method is a good tool for price forecasting as compared to other methods in terms of accuracy as well as convenience.
Short-Term Electricity Price Forecasting
Published in João P. S. Catalão, Electric Power Systems, 2017
Although time-series techniques usually are not taken into account as artificial intelligence-based forecasting engines, they are discussed here as these methods are used in many electricity price prediction research works. Among the time-series methods, autoregressive integrated moving average (ARIMA), dynamic regression, and transfer function models have gained more attention for electricity price prediction due to their easy implementation and relatively good performance. For instance, ARIMA, dynamic regression, and transfer function methods have been used for electricity price forecasting in several studies [14,28,30–33]. In the following, ARIMA time series and its variants are detailed. Dynamic regression and transfer function models have similar structures.
Machine Learning and Deep Learning for Intelligent and Smart Applications
Published in Mangesh M. Ghonge, Ramchandra Sharad Mangrulkar, Pradip M. Jawandhiya, Nitin Goje, Future Trends in 5G and 6G, 2021
High-cost volatility may have a direct effect on the stability of the smart grid electricity market. Thus, to prevent severe repercussions of market dynamics, efficient and precise price forecasts must be enforced. This chapter discusses two smart approaches using ML to tackle the issue of electricity price forecasting. First, to predict the hourly price, a Support Vector Regression model is used. Second, it will explore and compare the DL model with the Support Vector Regression model.
Electricity price forecasting using neural networks with an improved iterative training algorithm
Published in International Journal of Ambient Energy, 2018
Morteza Gholipour Khajeh, Akbar Maleki, Marc A. Rosen, Mohammad H. Ahmadi
The effectiveness of neural networks and efficient metaheuristic techniques has led to their application to modelling and optimisation problems in different areas (Toghyani et al. 2014; Ahmadi, Mehrpooya, and Khalilpoor 2016; Maleki, Hajinezhad, and Rosen 2016; Maleki, Khajeh, and Ameri 2016; Maleki, Khajeh, and Rosen 2016; Maleki, Pourfayaz, and Rosen 2016). Some other price forecasting methods have been put forward in recent years (Anbazhagan and Kumarappan 2013; Mandal et al. 2013; Yan and Chowdhury 2013; Shrivastava and Panigrahi 2014; Tripathi, Upadhyay, and Singh 2014; Wan et al. 2014; Chen and Chen 2015). A combination of a fuzzy inference system and least-squares estimation is proposed for prediction of the locational marginal prices (LMPs) in (Li et al. 2007). In Conejo et al. (2005), the electricity price series is decomposed by wavelet transforms with the aim of finding less volatile components and then each subseries is separately forecasted by the ARIMA technique. A combination of similar days and neural network techniques is proposed for LMP prediction in (Mandal et al. 2007). In García-Martos, Rodríguez, and Sánchez (2007), a mixed model is proposed for price forecasting, where weekends and weekdays are modelled by separate ARIMA time series techniques. A weighted nearest neighbours method is presented for electricity price forecasting in García-Martos, Rodríguez, and Sánchez (2007). In Ghasemi et al. (2016), a hybrid algorithm is proposed for simultaneously forecasting price and demand that uses a set of effective tools in the preprocessing portion, forecast engine and tuned algorithm. The proposed forecast algorithm can be viewed in terms of three main parts: Using a new Flexible Wavelet Packet Transform to decompose a signal into multiple terms at different frequencies, and a new feature selection method that employs Conditional Mutual Information and adjacent features in order to select valuable input data.Employing a novel Multi-Input Multi-Output model based on Nonlinear Least Square SVM (NLSSVM) and ARIMA in order to model the linear and nonlinear correlations between price and load in two stages.Using a modified version of the Artificial Bee Colony algorithm based on time-varying coefficients and a stumble generation operator in order to optimise NLSSVM parameters in a learning process. Despite the research performed in the area of electricity price forecasting, more accurate and robust price forecast methods are still required.
A new fuzzy logic based approach for optimal household appliance scheduling based on electricity price and load consumption prediction
Published in Advances in Building Energy Research, 2022
Sara Atef, Nourhan Ismail, Amr B. Eltawil
The electricity consumption prediction problem has been studied using various predictive methods, starting from statistics to machine learning-based approaches (Debnath & Mourshed, 2018). The main drawback of some of these methods was assuming a regular electricity consumption pattern, which is more predictable (Shi et al., 2018). However, in practice, the actual usage pattern dynamically changes over time, influenced by consumer behaviour. This behaviour is too stochastic to forecast because there are many factors involved, such as weather, occupancy rates, and thermal system qualifications. To tackle this issue, a recent research trend has been established which adopts Deep Learning (DL) techniques in load prediction and electricity price forecasting (Shi et al., 2018; Rahman et al., 2018; Mocanu et al., 2016; Suryanarayana et al., 2018; Zhang et al., 2015; Atef & Eltawil, 2019a, 2019b, 2020). In (Mocanu et al., 2016), two DL based models, namely Conditional Restricted Boltzmann Machine (CRBM) and Factored Conditional Restricted Boltzmann Machine (FCRBM), were used to predict the energy consumption with one-minute, 15-minutes, hourly, and weekly resolutions over a four-year period. The performance of the proposed models was compared with Artificial Neural Network (ANN), Support Vector Machine (SVM), and Recurrent Neural Network (RNN) based models over a one-year test period. The results showed that the CRBM and FCRBM had a better performance as the time horizon increased. Moreover, a novel pooling-based DL model was proposed in (Shi et al., 2018) with the objective to investigate household load forecasting model uncertainties using deep learning. The proposed model showed more accurate results in comparison with the Autoregressive Integrated Moving Average (ARIMA), Support Vector Regression (SVR), and RNN models. Furthermore, Rahman et al., (2018) introduced two deep RNN methods to forecast the electricity consumption over a medium-to-long term time horizon. In addition, they introduced an approach to handle missing data. The results indicated a lower relative error by the proposed models when compared with the classic Multi-Layered Perceptron (MLP) neural network. Furthermore, a Fuzzy Logic-based system was proposed to predict the electricity load consumption in (Bissey et al., 2017). The proposed approach included time, day, past electricity consumption, and temperature input variables. Although the efficiency of the Fuzzy Logic-based approach, it seemed to need extensive knowledge-based experience to construct the associated rule-base that included 2688 rules. Moreover, there is a difficulty in determining the appropriate membership functions’ numbers in order to improve the learning period. Therefore, a more suitable learning algorithm that requires less human interaction is needed. On the other hand, several methods were introduced to tackle the electricity price forecasting problem (Keles et al., 2016; Saâdaoui, 2017; Panapakidis & Dagoumas, 2016; Vijayalakshmi & Girish, 2015). However, using the DL approach has been investigated recently and showed to be an efficient tool that outperforms other traditional prediction techniques such as the Autoregressive (AR), Adaptive Neuro Fuzzy Inference System (ANFIS), and SVR models (Atef & Eltawil, 2019b; Lago et al., 2018).