A trend fixed on firstly and seasonal adjustment model combined with the ε-SVR for short-term forecasting of electricity demand

Short-term electricity demand forecasting has always been an essential instrument in power system planning and operation by which an electric utility plans and dispatches loading so as to meet system demand. The accuracy of the dispatching system, derived from the accuracy of demand forecasting and the forecasting algorithm used, will determines the economic of the power system operation as well as the stability of the whole society. This paper presents a combined ε-SVR model considering seasonal proportions based on development tendencies from history data. We use one-order moving averages to produce a comparatively smooth data series, taking the averaging period as the interval that can effectively eliminate the seasonal variation. We used the smoothed data series as the training set input for the ε-SVR model and obtained the corresponding forecasting value. Afterward, we accounted for the previously removed seasonal variation. As a case, we forecast northeast electricity demand of China using the new method. We demonstrated that this simple procedure has very satisfactory overall performance by an analysis of variance with relative verification and validation. Significant reductions in forecast errors were achieved.     

Prediction of daily peak electricity demand in South Africa using volatility forecasting models

Daily peak electricity demand forecasting in South Africa using a seasonal autoregressive integrated moving average (SARIMA) model, a SARIMA model with generalized autoregressive conditional heteroskedastic (SARIMA–GARCH) errors and a regression-SARIMA–GARCH (Reg-SARIMA–GARCH) model is presented in this paper. The GARCH modeling methodology is introduced to accommodate the possibility of serial correlation in volatility since the daily peak demand data exhibits non-constant mean and variance, and multiple seasonality corresponding to weekly and monthly periodicity. The proposed Reg-SARIMA–GARCH model is designed in such a way that the predictor variables are initially selected using a multivariate adaptive regression splines algorithm. The developed models are used for out of sample prediction of daily peak demand. A comparative analysis is done with a piecewise linear regression model. Results from the study show that the Reg-SARIMA–GARCH model produces better forecast accuracy with a mean absolute percent error (MAPE) of 1.42%.     

An integrated fuzzy regression algorithm for energy consumption estimation with non-stationary data: A case study of Iran

This study presents an integrated fuzzy regression and time series framework to estimate and predict electricity demand for seasonal and monthly changes in electricity consumption especially in developing countries such as China and Iran with non-stationary data. Furthermore, it is difficult to model uncertain behavior of energy consumption with only conventional fuzzy regression (FR) or time series and the integrated algorithm could be an ideal substitute for such cases. At First, preferred Time series model is selected from linear or nonlinear models. For this, after selecting preferred Auto Regression Moving Average (ARMA) model, Mcleod-Li test is applied to determine nonlinearity condition. When, nonlinearity condition is satisfied, the preferred nonlinear model is selected and defined as preferred time series model. At last, the preferred model from fuzzy regression and time series model is selected by the Granger-Newbold. Also, the impact of data preprocessing on the fuzzy regression performance is considered. Monthly electricity consumption of Iran from March 1994 to January 2005 is considered as the case of this study. The superiority of the proposed algorithm is shown by comparing its results with other intelligent tools such as Genetic Algorithm (GA) and Artificial Neural Network (ANN).     

Monthly natural gas demand forecasting by adjusted seasonal grey forecasting model

ABSTRACT

Natural gas stands out among fossil fuels because it is relatively cleaner. It is also an important energy source type for several fields such as electricity production, industry, and heating, etc. Due to the poor capacity of Turkey in terms of natural gas sources, the demand is supplied by producer countries. Hence, accurate forecasting for the demand is of critical importance for Turkey, which imports 99% of its natural gas consumption. In the current literature about demand forecasting, most studies were conducted on an annual basis. However, the seasonal effect on the demand for natural gas cannot be foreseen through annual studies. Besides, to deal with some situations such as seasonal balancing, peak shaving, and gas supply shortage in monthly demand, forecasting models that capture the seasonal trend are needed. Therefore, in this study, a new grey seasonal forecast model has been presented and Turkey’s monthly natural gas demand was predicted via the proposed model. Performance of that model was compared with SGM(1,1) and SARIMA (p,d,q) x (P,D,Q)^s. The obtained results show the superiority of the proposed model. By using this model, Turkey’s monthly natural gas demand was forecasted up until the year 2025. The proposed model allows us to capture seasonal patterns more successfully. In case this seasonal behavior continues, Turkey’s natural gas demand is expected to increase by %20 until 2025. At this point, the outcomes of the study provide important information to decision-makers to be able to determine reliable and stable energy policies.     

A hybrid short-term electricity price forecasting framework: Cuckoo search-based feature selection with singular spectrum analysis and SVM

Under the liberalization and deregulation of the power industry, price forecasting has become a cornerstone for market participants' decision-making such as bidding strategies and purchase plans. However, the exclusive nonlinearity dynamics of electricity price is a challenge problem that largely affects forecasting accuracy. To address this task, this paper presents a hybrid forecasting framework for short-term electricity price forecasting by exploiting and mining the important information hidden in the electricity price signal. Moreover, a hybrid feature selection method (HFS) is introduced into the forecasting strategy. To exhibit the dynamical characteristics of electricity price, we primarily perform a singular spectrum analysis (SSA)-based systematic analysis process by using the merit of SSA and analyzing the multiple seasonal patterns of short-term electricity price series, providing a meaningful representation of the hidden patterns and time-varying volatility of electricity price series. Aiming at selecting the key features, the candidate variables are constructed considering the dynamic behavior of price series; further, to capture the optimal features from the candidates, the correlation threshold θ is defined for the adjustable parameters in HFS and optimally determined by the intelligent search algorithm. Additionally, triangulation based on the Pearson, Spearman and Kendall rank correlation coefficient is performed to strengthen the reliability of the proposed method. The proposed hybrid forecasting framework is validated in the New South Wales electricity market, which demonstrates that the developed approach is truly better than the benchmark models used and a reliable and promising tool for short-term electricity price forecasting.     

Correlated synthetic time series generation for energy system simulations using Fourier and ARMA signal processing

As the contribution of renewable energy grows in electricity markets, the complexity of the energy mix required to meet demand grows, likewise the need for robust simulation techniques. While decades of wind, solar, and demand profiles can sometimes be obtained, this is too few samples to provide a statistically meaningful analysis of a system with baseload, peaker, and renewable generation. To demonstrate the viability of an energy mix, many thousands of samples are needed. Synthetic time series generation presents itself as a suitable methodology to meet this need. For a synthetic time series to be statistically viable, several conditions must be met. The series generator must produce independent, identically distributed samples, each having the same fundamental properties as the original signal without duplicating it exactly. One approach for such a generator is training a surrogate model using Fourier series decomposition for seasonal patterns and autoregressive moving average (ARMA) models to describe time-correlated statistical noise about the seasonal patterns. When combined, the Fourier plus ARMA (FARMA) model has been shown to provide an infinite set of independent, identically distributed sample time series with the same statistical properties as the original data [1]. When considering an energy mix with renewable electricity production, several time series of energy, grid, and weather measurements are needed for each synthetic year modeled to statistically comprehend the efficiency of any given energy mix. This includes measurements of solar exposure, air temperature, wind velocity, and electricity demand. These cannot be considered independent series in a given synthetic year; for example, in summer months demand may be higher when solar exposure and air temperature are high and wind velocity is low. To capture and reproduce the correlations that might exist in the measured histories, the ARMA can further be extended as a Vector ARMA (VARMA). In the VARMA algorithm, covariance in statistical noise is captured both within a history as part of the autoregressive moving average and with respect to the other variables in the time series. In this work, the implementation of the Fourier VARMA in the RAVEN uncertainty quantification and risk analysis software framework [2] is presented, along with examples of correlated synthetic history generation. Finally, methods to extend synthetic signals to multiyear samples are presented and discussed.     

Prediction of wind speed time series using modified Taylor Kriging method

Wind speed forecasting is critical for the operations of wind turbine and penetration of wind energy into electricity systems. In this paper, a novel time series forecasting method is proposed for this purpose. This method originates from TK (Taylor Kriging) model, but is properly modified for the forecasting of wind speed time series. To investigate the performance of this new method, the wind speed data from an observation site in North Dakota, USA, are adopted. One-year hourly wind speed data are divided into 10 samples, and forecast is made for each sample. In the case study, both the modified TK method and (ARIMA) autoregressive integrated moving average method are employed and their performances are compared. It is found that on average, the proposed method outperforms the ARIMA method by 18.60% and 15.23% in terms of (MAE) mean absolute error and (RMSE) root mean square error. Meanwhile, further theoretical analysis is provided to discuss why the modified TK method is potentially more accurate than the ARIMA method for wind speed time series prediction.     

基于集合经验模态分解和自回归滑动平均的某碾压混凝土重力坝变形预测模型及应用

为更精确地预测大坝变形数据,针对大坝变形监测序列的非线性和非平稳性问题,提出了一种结合集合经验模态分解和自回归滑动平均模型的大坝变形预测模型。首先利用集合经验模态分解法将非平稳的大坝变形监测数据分解为具有不同特征尺度的本征模态函数,然后分析各分量特征并分别建立自回归滑动平均模型,选择各自适合的最优模型参数,最后叠加各分量的预测结果作为最终的变形预测结果。分析结果表明,相较单一预测模型,结合集合经验模态分解和自回归滑动平均模型的组合预测模型的预测精度更高。     

Forecasting electricity price and demand using a hybrid approach based on wavelet transform,ARIMA and neural networks

Demand and price forecasting are extremely important for participants in energy markets. Most research work in the area predicts demand and price signals separately. In this paper, a model is presented which predicts electricity demand and price simultaneously. The model combines wavelet transforms, ARIMA models and neural networks. Both time domain and wavelet domain variables are considered in the feature set for price and demand forecasting. The best input set is selected by two‐step correlation analysis. The proposed model is better adapted to real conditions of an energy market since the forecast features for price and demand are not assumed as known values but are predicted by the model, thus accounting for the interactions of the demand and price forecast processes. The forecast accuracy of the proposed method is evaluated using data from the Finnish energy market, which is part of the Nordic Power Pool. The results show that the proposed model provides significant improvement in both demand and price prediction accuracy compared with models using a separate frameworks approach. Copyright © 2013 John Wiley & Sons, Ltd.     

Electricity price forecasting using sale and purchase curves: The X-Model

Our paper aims to model and forecast the electricity price by taking a completely new perspective on the data. It will be the first approach which is able to combine the insights of market structure models with extensive and modern econometric analysis. Instead of directly modeling the electricity price as it is usually done in time series or data mining approaches, we model and utilize its true source: the sale and purchase curves of the electricity exchange. We will refer to this new model as X-Model, as almost every deregulated electricity price is simply the result of the intersection of the electricity supply and demand curve at a certain auction. Therefore we show an approach to deal with a tremendous amount of auction data, using a subtle data processing technique as well as dimension reduction and lasso based estimation methods. We incorporate not only several known features, such as seasonal behavior or the impact of other processes like renewable energy, but also completely new elaborated stylized facts of the bidding structure. Our model is able to capture the non-linear behavior of the electricity price, which is especially useful for predicting huge price spikes. Using simulation methods we show how to derive prediction intervals for probabilistic forecasting. We describe and show the proposed methods for the day-ahead EPEX spot price of Germany and Austria.     

A novel seasonal decomposition based least squares support vector regression ensemble learning approach for hydropower consumption forecasting in China

Due to the distinct seasonal characteristics of hydropower, this study tries to propose a seasonal decomposition (SD) based least squares support vector regression (LSSVR) ensemble learning model for Chinese hydropower consumption forecasting. In the formulation of ensemble learning model, the original hydropower consumption series are first decomposed into trend cycle, seasonal factor and irregular component. Then the LSSVR with the radial basis function (RBF) kernel is used to predict the three different components independently. Finally, these prediction results of the three components are combined with another LSSVR to formulate an ensemble result for the original hydropower consumption series. In terms of error measurements and statistic test on the forecasting performance, the proposed approach outperforms all the other benchmark methods listed in this study in both level accuracy and directional accuracy. Experimental results reveal that the proposed SD-based LSSVR ensemble learning paradigm is a very promising approach for complex time series forecasting with seasonality.     

Data analysis and short term load forecasting in Iran electricity market using singular spectral analysis (SSA)

In this paper, the data analysis and short term load forecasting (STLF) in Iran electricity market has been considered. The proposed method is an improved singular spectral analysis (SSA) method. SSA decomposes a time series into its principal components i.e. its trend and oscillation components, which are then used for time series forecasting, effectively. The employed data are the total load time series of Iran electricity market in its real size and is long enough to make it possible to take properties such as non-stationary and annual periodicity of the market into account. Simulation results show that the proposed method has a good ability in characterizing and prediction of the desired load time series in comparison with some other related methods.     

The application of seasonal latent variable in forecasting electricity demand as an alternative method

In this study, we used ARIMA, seasonal ARIMA (SARIMA) and alternatively the regression model with seasonal latent variable in forecasting electricity demand by using data that belongs to “Kayseri and Vicinity Electricity Joint-Stock Company” over the 1997:1–2005:12 periods. This study tries to examine the advantages of forecasting with ARIMA, SARIMA methods and with the model has seasonal latent variable to each other. The results support that ARIMA and SARIMA models are unsuccessful in forecasting electricity demand. The regression model with seasonal latent variable used in this study gives more successful results than ARIMA and SARIMA models because also this model can consider seasonal fluctuations and structural breaks.     

Frequency domain methods applied to forecasting electricity markets

The changes taking place in electricity markets during the last two decades have produced an increased interest in the problem of forecasting, either load demand or prices. Many forecasting methodologies are available in the literature nowadays with mixed conclusions about which method is most convenient. This paper focuses on the modeling of electricity market time series sampled hourly in order to produce short-term (1 to 24 h ahead) forecasts. The main features of the system are that (i) models are of an Unobserved Component class that allow for signal extraction of trend, diurnal, weekly and irregular components; (ii) its application is automatic, in the sense that there is no need for human intervention via any sort of identification stage; (iii) the models are estimated in the frequency domain; and (iv) the robustness of the method makes possible its direct use on both load demand and price time series. The approach is thoroughly tested on the PJM interconnection market and the results improve on classical ARIMA models.     

On the importance of the long-term seasonal component in day-ahead electricity price forecasting

In day-ahead electricity price forecasting (EPF) the daily and weekly seasonalities are always taken into account, but the long-term seasonal component (LTSC) is believed to add unnecessary complexity to the already parameter-rich models and is generally ignored. Conducting an extensive empirical study involving state-of-the-art time series models we show that (i) decomposing a series of electricity prices into a LTSC and a stochastic component, (ii) modeling them independently and (iii) combining their forecasts can bring – contrary to a common belief – an accuracy gain compared to an approach in which a given time series model is calibrated to the prices themselves.     

Generation of synthetic sequences of electricity demand: Application in South Australia

We have developed a model to generate synthetic sequences of half-hourly electricity demand. The generated sequences represent possible realisations of electricity load that could have occurred. Each of the components included in the model has a physical interpretation. These components are yearly and daily seasonality which were modelled using Fourier series, weekly seasonality modelled with dummy variables, and the relationship with current temperature described by polynomial functions of temperature. Finally the stochastic component was modelled with autoregressive moving average (ARMA) processes. These synthetic sequences were developed for two purposes. The first one is to use them as input data in market simulation software. The second one is to build probability distributions of the outputs to calculate probabilistic forecasts. As an application several summers of half-hourly electricity demand were generated and from them the value of demand that is not expected to be exceeded more than once in 10 years was calculated.     

An effective and robust decomposition-ensemble energy price forecasting paradigm with local linear prediction

Energy price time series exhibit nonlinear and nonstationary features, which make accurate forecasting energy prices challenging. In this paper, we propose a novel decomposition-ensemble forecasting paradigm based on ensemble empirical mode decomposition (EEMD) and local linear prediction (LLP). The EEMD is used to decompose energy price time series into components, including several intrinsic mode functions and one residual with a simplified structure. Motivated by the findings of the fully local characteristics of a time series decomposed by the EEMD, we adopt the LLP technique to forecast each component. The forecasting results of all the components are aggregated as a final forecast. For validation, three types of energy price time series, crude oil, electricity and natural gas prices, are studied. The experimental results indicate that the proposed model achieves an improvement in terms of both level forecasting and direction forecasting. The performance of the proposed model is also validated through comparison with several energy price forecasting approaches from the literature. In addition, the robustness and the effects of the parameter settings of LLP are investigated. We conclude the proposed model is easy to implement and efficient for energy price forecasting.     

Machine learning based switching model for electricity load forecasting

In deregulated power markets, forecasting electricity loads is one of the most essential tasks for system planning, operation and decision making. Based on an integration of two machine learning techniques: Bayesian clustering by dynamics (BCD) and support vector regression (SVR), this paper proposes a novel forecasting model for day ahead electricity load forecasting. The proposed model adopts an integrated architecture to handle the non-stationarity of time series. Firstly, a BCD classifier is applied to cluster the input data set into several subsets by the dynamics of the time series in an unsupervised manner. Then, groups of SVRs are used to fit the training data of each subset in a supervised way. The effectiveness of the proposed model is demonstrated with actual data taken from the New York ISO and the Western Farmers Electric Cooperative in Oklahoma.     

Integration of genetic algorithm,computer simulation and design of experiments for forecasting electrical energy consumption

This study presents an integrated algorithm for forecasting monthly electrical energy consumption based on genetic algorithm (GA), computer simulation and design of experiments using stochastic procedures. First, time-series model is developed as a benchmark for GA and simulation. Computer simulation is developed to generate random variables for monthly electricity consumption. This is achieved to foresee the effects of probabilistic distribution on monthly electricity consumption. The GA and simulated-based GA models are then developed by the selected time-series model. Therefore, there are four treatments to be considered in analysis of variance (ANOVA) which are actual data, time series, GA and simulated-based GA. Furthermore, ANOVA is used to test the null hypothesis of the above four alternatives being equal. If the null hypothesis is accepted, then the lowest mean absolute percentage error (MAPE) value is used to select the best model, otherwise the Duncan Multiple Range Test (DMRT) method of paired comparison is used to select the optimum model, which could be time series, GA or simulated-based GA. In case of ties the lowest MAPE value is considered as the benchmark. The integrated algorithm has several unique features. First, it is flexible and identifies the best model based on the results of ANOVA and MAPE, whereas previous studies consider the best-fit GA model based on MAPE or relative error results. Second, the proposed algorithm may identify conventional time series as the best model for future electricity consumption forecasting because of its dynamic structure, whereas previous studies assume that GA always provide the best solutions and estimation. To show the applicability and superiority of the proposed algorithm, the monthly electricity consumption in Iran from March 1994 to February 2005 (131 months) is used and applied to the proposed algorithm.     

城市用水量组合预测模型及其应用

针对城市用水量时间序列包含逐步增长趋势、季节性趋势及不确定性的非线性波动特点,单一预测模型往往很难充分反映原始数据中全部的有效信息,结合季节性时间序列模型(SARIMA)和BP神经网络二者优点,构建了一种新型的组合预测模型,对上海市用水量进行不同时间尺度的预测。结果表明,在不同时间尺度上组合预测模型均比单一预测模型精度高、预测质量稳定。