No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Assessment of some methods for short-term load forecasting
Abstract
Accurate forecasting of loads is essential for smart grids and energy markets. This paper compares the performance of the following models in short-term load forecasting: 1) smart metering data based profile models, 2) a neural network (NN) model, and 3) a Kalman-filter based predictor with input nonlinearities and a physically based main structure. The comparison helps method selection for the development of hybrid models for forecasting the load control responses. According to the results all these three modeling approaches show much better performance than 4) the traditional load profiles and 5) a static outdoor temperature dependency model applied with a lag. The neural network model was the most accurate in the comparison, but the differences of the three methods developed were rather small and also other aspects and other methods must be considered and compared when selecting the method for a specific purpose.
... In addition to cluster profiles, individual load profiles are calculated for those large customers that exhibit unique consumption characteristics. Seasonal temperature dependencies are calculated for every cluster and individual profile and when combined with outdoor temperature forecasts, the new load profiles can be used also for load forecasting [4]. ...
... During the SGEM project, several different models utilizing hourly metered consumption data were evaluated for shortterm load forecasting. The studied models were; a cluster profile based predictor, a Kalman-filter based predictor with input nonlinearities and physically based main structure, a neural network (NN) model [4], and a support vector machine (SVM) model [8]. The NN and SVM models were the most accurate, but also the other methods had their relative merits. ...
Smart meters collect a lot of data on customer level electricity consumption and this, together with other data sources e.g. environmental information and public open data, provides an excellent basis for data mining. As a part of a recent smart grid project conducted in Finland, several different ways of mining smart meter data were studied. The project brought advances in customer classification and clustering, load profiling, spatial load analytics, behaviour change detection and load forecasting.
... They compared linear regression, support vector regression (SVR) and multilayer perceptron (MLP) in this respect. [26] shows a typical short-term load forecasting accuracy dependence on the prediction time horizon. The weather forecasts and load forecasting methods have improved much so now the accuracy decreases somewhat later but the shape of the dependency is still similar. ...
When identifying and comparing forecasting models, there may be a risk that poorly selected criteria could lead to wrong conclusions. Thus, it is important to know how sensitive the results are to the selection of criteria. This contribution aims to study the sensitivity of the identification and comparison results to the choice of criteria. It compares typically applied criteria for tuning and performance assessment of load forecasting methods with estimated costs caused by the forecasting errors. The focus is on short-term forecasting of the loads of energy systems. The estimated costs comprise electricity market costs and network costs. We estimate the electricity market costs by assuming that the forecasting errors cause balancing errors and consequently balancing costs to the market actors. The forecasting errors cause network costs by overloading network components thus increasing losses and reducing the component lifetime or alternatively increase operational margins to avoid those overloads. The lifetime loss of insulators, and thus also the components, is caused by heating according to the law of Arrhenius. We also study consumer costs. The results support the assumption that there is a need to develop and use additional and case-specific performance criteria for electricity load forecasting.
... Kalman filtering is presented to forecast short term load [16]. These models cannot handle a large amount of data and do not deliver good accuracy [17]. The review of different methods for forecasting is shown in Table VI. ...
... It develops smart metering based load profiles to support state estimation of the electricity distribution system [7]. In our initial study [8], the smart meter based load profiles showed roughly equal performance with two other short-term load forecasting methods that were a neural network, and a Kalman-filter based predictor with input nonlinearities and a physically based main structure. These three methods had the best performance in the study. ...
... Reference [2] compares the performance of smart metering data-based profile models; a Kalman filter-based predictor with input nonlinearities and a neural network (NN) model for the STLF problem. An accurate STLF model using discrete wavelet transform in combination with artificial neural network/support vector machine is proposed in [3]. ...
This paper proposes a Bat algorithm-based back propagation approach for solving the short-term load forecasting considering the weather factors such as temperature and humidity. The accuracy of load forecasting is very important in the restructured power system as it plays a major role in providing a better cost-effective risk management and operation plans. Load/demand forecasting is difficult due to the nonlinear and randomness properties load itself, its dependency on factors like weather conditions, variations of social and economic environments, and price of the electrical power in deregulated environment. Load forecasting accuracy significantly impacts the cost of power utilities in operational planning of the energy supply. To show the effectiveness of the proposed approach, the PJM (Pennsylvania–New Jersey–Maryland) system load demand data are considered. The simulation results obtained have shown that the day-ahead hourly forecasts of load demand using the proposed method is very accurate with very less error and well forecasted.
In traditional power networks, Distribution System Operators (DSOs) used to monitor energy flows on a medium- or high-voltage level for an ensemble of consumers and the low-voltage grid was regarded as a black box. However, electric utilities nowadays obtain ever more precise information from single consumers connected to the low- and medium-voltage grid thanks to smart meters (SMs). This allows a previously unattainable degree of detail in state estimation and other grid analysis functionalities such as predictions. This paper focuses on the use of Artificial Neural Networks (ANNs) for accurate short-term load and Photovoltaic (PV) predictions of SM profiles and investigates different spatial aggregation levels. A concluding power flow analysis confirms the benefits of time series prediction to support grid operation. This study is based on the SM data available from more than 40,000 consumers as well as PV systems in the City of Basel, Switzerland.
Accurate electrical load forecasts are of vital interest to power companies. Short term load forecasts for next hours in particular are important for power dispatch, power trading and system operation. This paper analyzes the conjectures that a self-adaptive weighting algorithm (SAW), blending different standard load forecasting approaches, such as a dynamic standard load profile model, a linear regression model and an artificial neuronal network model, can increase forecasting performance on micro grids for one hour intraday to 24 hours day ahead forecasts. The SAW methodology and forecasting models are applied to a publicly available smart meter data set. Common evaluation metrics such as the mean average percentage error (MAPE) and the normalized rooted mean square error (NRMSE) are used to evaluate the performance of this new hybrid approach and allow a comparison to other studies. Self-adaptive weighing leads to a significant improvement of intraday and day ahead forecasts from 50%–54% and 30%–35% (MAPE improvement for 1h and 24h compared to input forecasts). The resulting intraday and day ahead load SAW forecasts range from 3.19% 1h MAPE to 4.50% 24h MAPE in this case study.
Short-term forecasting of electric loads is an essential function required by Smart Grids. Today increasing
amount of smart metering data is available enabling the development of enhanced data-driven models for short-term load forecasting. Until now, a plethora of models have been developed ranging from simple linear regression models to more advanced models such as (artificial) neural networks (NNs) and support
vector machines (SVMs). Despite the relatively high accuracy obtained, the acceptance of purely data-driven models such as NN models is still remained limited due to their complexity and nontransparent nature. Therefore it is important to develop optimization schemes, which can be used to facilitate the selection
of appropriate model structure resulting good forecasting accuracy with low complexity. This study presents an optimization scheme based on multi-objective genetic algorithm (GA) for designing data-driven models for short-term forecasting of electric loads. The optimization scheme is demonstrated for designing the
conventional NN/MLP model using real smart metering data and weather measurements. The optimal NN model structures are identified and analyzed in terms of model complexity and forecasting accuracy.
In Finland, customer class load profiles are used extensively in distribution network calculation. State estimation systems, for example, use the load profiles to estimate the state of the network. Load profiles are also needed to predict future loads in distribution network planning. In general, customer class load profiles are obtained through sampling in load research projects. Currently, in Finland, customer classification is based on the uncertain customer information found in the customer information system. Customer information, such as customer type, heating solution, and tariff, is used to connect the customers with corresponding customer class load profiles. Now that the automatic meter-reading systems are becoming more common, customer classification and load profiling could be done according to actual consumption data. This paper proposes the use of the ISODATA algorithm for customer classification. The proposed customer classification and load profiling method also includes temperature dependency correction and outlier filtering. The method is demonstrated in this paper by studying a set of 660 hourly metered customers.
The deregulation of electric power supply industries has raised many challenging problems. One of the most important ones is forecasting the Market Clearing Price (MCP) of electricity. Decisions on various issues, such as to buy or sell electricity and to offer a transaction to the market, require accurate knowledge of the MCP. Another problem, which has also been an important issue of the traditional power systems, is load forecasting for both short and long terms. The extended kalman filter has been widely adopted for state estimation of nonlinear systems, machine learning applications and neural network training. In the EKF, the state distribution is approximated by the first-order linearization of the nonlinear system. Therefore this can introduce large errors in the load and price forecasting as two Chaotic, nonstationary and nonlinear time-series. The unscented Kalman filter (UKF), in contrast, achieves third-order accuracy, by using a minimal set of MCP and load sigma points. In this paper an improved dual unscented Kalman filter (DUKF), which estimate state and parameter simultaneously has been applied to the real New England power market. The numerical stability and more accurate predictions of our method is comparable to the EKF, and traditional neural network training methods. Remarkably, the computational complexity of the DUKF is the same order as that of the EKE. The obtained results show significant improvement in both price and load forecasting.
The topics introduced in this thesis are: the Finnish load research project, a simple form customer class load model, analysis of the origins of customer's load distribution, a method for the estimation of the confidence interval of customer loads and Distribution Load Estimation (DLE) which utilises both the load models and measurements from distribution networks. These developments bring new knowledge and understanding of electricity customer loads, their statistical behaviour and new simple methods of how the loads should be estimated in electric utility applications. The economic benefit is to decrease investment costs by reducing the planning margin when the loads are more reliably estimated in electrc utilities. As the Finnish electricity production, transmission and distribution is moving towards the de-regulated electricity markets, this study also contributes to the development for this new situation. The Finnish load research project started in 1983. The project was initially coordinated by the Association of Finnish Electric Utilities and 40 utilities joined the project. Now there are over 1000 customer hourly load recordings in a database. A simple form customer class load model is introduced. The model is designed to be practical for most utility applications and has been used by the Finnish utilities for several years. There is now available models for 46 different customer classes. The only variable of the model is the customer's annual energy consumption. The model gives the customer's average hourly load and standard deviation for a selected month, day and hour. The statistical distribution of customer loads is studied and a model for customer electric load variation is developed. The model results in a lognormal distribution as an extreme case. The model is easy to simulate and produces distributions similar to those observed in load research data. Analysis of the load variation model is an introduction to the further analysis of methods for confidence interval estimation. Using the 'simple form load model', a method for estimating confidence intervals (confidence limits) of customer hourly load is developed. The two methods selected for final analysis are based on normal and lognormal distribution estimated in a simplified manner. The simplified lognormal estimation method is a new method presented in this thesis. The estimation of several cumulated customer class loads is also analysed. Customer class load estimation which combines the information from load models and distribution network load measurements is developed. This method, called Distribution Load Estimation (DLE), utilises information already available in the utility's databases and is thus easy to apply. The resulting load data is more reliable than the load models alone. One important result of DLE is the estimate of the customer class' share to the distribution system's total load.
Automatic meter reading (AMR) is becoming common in many European countries. This paper shows how AMR measurements can be used to create new load profiles and how these new load profiles can be applied to improve distribution network analysis accuracy. In this paper, hourly electricity consumption data is used to update existing load profiles, cluster customers and create new cluster profiles, and specify individual profiles for selected customers, all of which are then used in distribution network analysis. The results between existing and new load profiling methods are compared. Comparisons are also made between different methods of AMR-based load profiling.
Recent researches in load forecasting are quite often based on the use of neural networks in order to predict a specific variable (maximum demand, active electric power or hourly consumption) using past values of the same variable and other exogenous factors proved to influence the value being predicted. This work aims to explore different input patterns in neural networks incorporating information derived from load profiles of different consumers' classes.
For decision makers in the electricity sector, the decision process is complex with several different levels that have to be taken into consideration. These comprise for instance the planning of facilities and an optimal day-to-day operation of the power plant. These decisions address widely different time-horizons and aspects of the system. For accomplishing these tasks load forecasts are very important. Therefore, finding an appropriate approach and model is at core of the decision process. Due to the deregulation of energy markets, load forecasting has gained even more importance. In this article, we give an overview over the various models and methods used to predict future load demands.
We discuss and compare measures of accuracy of univariate time series forecasts. The methods used in the M-competition as well as the W-competition, and many of the measures recommended by previous authors on this topic, are found to be degenerate in commonly occurring situations. Instead, we propose that the mean absolute scaled error become the standard measure for comparing forecast accuracy across multiple time series. (c) 2006 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.
A review of five widely applied short-term (up to 24 h) load
forecasting techniques is presented. These are: multiple linear
regression; stochastic time series; general exponential smoothing; state
space and Kalman filter; and a knowledge-based approach. A brief
discussion of each of these techniques, along with the necessary
equations, is presented. Algorithms implementing these forecasting
techniques have been programmed and applied to the same database for
direct comparison of these different techniques. A comparative summary
of the results is presented to give an understanding of the inherent
level of difficulty of each of these techniques and their performances
Measurements and models of electricity demand responses
- P Koponen
Load and response modelling workshop in project SGEM. 10 November 2011, Kuopio. Espoo, VTT, VTT Working Papers 188
- P Koponen
- J Saarenpää