WEATHER DEPENENT ELECTRICITY MARKET FORECASTING WITH NEURAL NETWORKS, WAVELET AND DATA MINING TECHNIQUES. Z.Y. Dong X. Li Z. Xu K. L.

Similar documents
Demand Forecasting in Deregulated Electricity Markets

A Hybrid Model of Wavelet and Neural Network for Short Term Load Forecasting

One-Hour-Ahead Load Forecasting Using Neural Network

Electric Load Forecasting Using Wavelet Transform and Extreme Learning Machine

Day Ahead Hourly Load and Price Forecast in ISO New England Market using ANN

A new method for short-term load forecasting based on chaotic time series and neural network

HYBRID ARTIFICIAL NEURAL NETWORK SYSTEM FOR SHORT-TERM LOAD FORECASTING

Short Term Load Forecasting Using Multi Layer Perceptron

Forecasting demand in the National Electricity Market. October 2017

Neural-wavelet Methodology for Load Forecasting

Short Term Load Forecasting Based Artificial Neural Network

MODELLING ENERGY DEMAND FORECASTING USING NEURAL NETWORKS WITH UNIVARIATE TIME SERIES

Load Forecasting Using Artificial Neural Networks and Support Vector Regression

Univariate versus Multivariate Models for Short-term Electricity Load Forecasting

LONG - TERM INDUSTRIAL LOAD FORECASTING AND PLANNING USING NEURAL NETWORKS TECHNIQUE AND FUZZY INFERENCE METHOD ABSTRACT

Implementation of Artificial Neural Network for Short Term Load Forecasting

Energy Clearing Price Prediction and Confidence Interval Estimation With Cascaded Neural Networks

USE OF FUZZY LOGIC TO INVESTIGATE WEATHER PARAMETER IMPACT ON ELECTRICAL LOAD BASED ON SHORT TERM FORECASTING

Evaluation of support vector machine based forecasting tool in electricity price forecasting for Australian national electricity market participants *

ANN and Statistical Theory Based Forecasting and Analysis of Power System Variables

Fuzzy based Day Ahead Prediction of Electric Load using Mahalanobis Distance

A Hybrid Method of CART and Artificial Neural Network for Short-term term Load Forecasting in Power Systems

Short-Term Load Forecasting Using ARIMA Model For Karnataka State Electrical Load

Short Term Load Forecasting for Bakhtar Region Electric Co. Using Multi Layer Perceptron and Fuzzy Inference systems

Influence of knn-based Load Forecasting Errors on Optimal Energy Production

Short-term load forecasting in large scale electrical utility using artificial neural network

A Fuzzy Logic Based Short Term Load Forecast for the Holidays

COMPARISON OF CLEAR-SKY MODELS FOR EVALUATING SOLAR FORECASTING SKILL

A data mining approach for medium-term demand forecasting

A Feature Based Neural Network Model for Weather Forecasting

Electricity price forecasting in Turkey with artificial neural network models

Stream-Based Electricity Load Forecast

Explanatory Information Analysis for Day-Ahead Price Forecasting in the Iberian Electricity Market

MODELLING TRAFFIC FLOW ON MOTORWAYS: A HYBRID MACROSCOPIC APPROACH

Forecasting Crude Oil Price Using Neural Networks

Forecasting electricity market pricing using artificial neural networks

Direct Method for Training Feed-forward Neural Networks using Batch Extended Kalman Filter for Multi- Step-Ahead Predictions

Short-term water demand forecast based on deep neural network ABSTRACT

A Unified Framework for Near-term and Short-term System Load Forecasting

Journal of of Computer Applications Research Research and Development and Development (JCARD), ISSN (Print), ISSN

Application of Artificial Neural Network for Short Term Load Forecasting

FORECASTING SAVING DEPOSIT IN MALAYSIAN ISLAMIC BANKING: COMPARISON BETWEEN ARTIFICIAL NEURAL NETWORK AND ARIMA

LONG TERM LOAD FORECASTING OF POWER SYSTEMS USING ARTIFICIAL NEURAL NETWORK AND ANFIS

Short Term Load Forecasting Of Chhattisgarh Grid Using Artificial Neural Network

A Wavelet Neural Network Forecasting Model Based On ARIMA

Weighted Fuzzy Time Series Model for Load Forecasting

Intelligent Modular Neural Network for Dynamic System Parameter Estimation

NEURAL NETWORK BASED APPROACH FOR SHORT-TERM LOAD FORECASTING

Artificial Neural Network for Energy Demand Forecast

WIND INTEGRATION IN ELECTRICITY GRIDS WORK PACKAGE 3: SIMULATION USING HISTORICAL WIND DATA

Wavelet Neural Networks for Nonlinear Time Series Analysis

LONG TERM LOAD FORECASTING FOR THE EGYPTIAN NETWORK USING ANN AND REGRESSION MODELS

Artificial Neural Network-Based Short-Term Demand Forecaster

FORECASTING OF ECONOMIC QUANTITIES USING FUZZY AUTOREGRESSIVE MODEL AND FUZZY NEURAL NETWORK

Australian Journal of Basic and Applied Sciences. A Comparative Analysis of Neural Network based Short Term Load Forecast for Seasonal Prediction

Predicting the Electricity Demand Response via Data-driven Inverse Optimization

Fuzzy Logic Approach for Short Term Electrical Load Forecasting

Open Access Research on Data Processing Method of High Altitude Meteorological Parameters Based on Neural Network

Spatially-Explicit Prediction of Wholesale Electricity Prices

Data and prognosis for renewable energy

RESERVOIR INFLOW FORECASTING USING NEURAL NETWORKS

This paper presents the

LOAD FORECASTING APPLICATIONS for THE ENERGY SECTOR

CHAPTER 6 CONCLUSION AND FUTURE SCOPE

CHAPTER 5 - QUEENSLAND FORECASTS

Heat Load Forecasting of District Heating System Based on Numerical Weather Prediction Model

Segmenting electrical load time series for forecasting? An empirical evaluation of daily UK load patterns

AESO Load Forecast Application for Demand Side Participation. Eligibility Working Group September 26, 2017

Markovian Models for Electrical Load Prediction in Smart Buildings

Introduction to Course

Neural Networks for Short Term Wind Speed Prediction

Journal of Chemical and Pharmaceutical Research, 2014, 6(5): Research Article

Probabilistic Energy Forecasting

Longshore current velocities prediction: using a neural networks approach

Optimum Neural Network Architecture for Precipitation Prediction of Myanmar

ACCURATE load forecasting is a key requirement for the

Lecture Prepared By: Mohammad Kamrul Arefin Lecturer, School of Business, North South University

A Bayesian Perspective on Residential Demand Response Using Smart Meter Data

An efficient approach for short-term load forecasting using historical data

Forecasting of Electric Consumption in a Semiconductor Plant using Time Series Methods

MURDOCH RESEARCH REPOSITORY

Electric Load Forecasting: An Interval Type-II Fuzzy Inference System Based Approach

Distributed Clustering and Local Regression for Knowledge Discovery in Multiple Spatial Databases

FORECAST ACCURACY REPORT 2017 FOR THE 2016 NATIONAL ELECTRICITY FORECASTING REPORT

IJESRT. Scientific Journal Impact Factor: (ISRA), Impact Factor: [Gupta et al., 3(5): May, 2014] ISSN:

Do we need Experts for Time Series Forecasting?

22/04/2014. Economic Research

A Sparse Linear Model and Significance Test. for Individual Consumption Prediction

Short Term Load Forecasting via a Hierarchical Neural Model

A Dynamic Combination and Selection Approach to Demand Forecasting

Multivariate Regression Model Results

Neural Networks Introduction

On the importance of the long-term seasonal component in day-ahead electricity price forecasting. Part II - Probabilistic forecasting

AUTO SALES FORECASTING FOR PRODUCTION PLANNING AT FORD

Evolutionary Functional Link Interval Type-2 Fuzzy Neural System for Exchange Rate Prediction

A Wavelet Based Prediction Method for Time Series

A Hybrid ARIMA and Neural Network Model to Forecast Particulate. Matter Concentration in Changsha, China

Wind Power Forecasting using Artificial Neural Networks

FORECASTING YIELD PER HECTARE OF RICE IN ANDHRA PRADESH

Are Daily and Weekly Load and Spot Price Dynamics in Australia s National Electricity Market Governed by Episodic Nonlinearity?

Transcription:

WEATHER DEPENENT ELECTRICITY MARKET FORECASTING WITH NEURAL NETWORKS, WAVELET AND DATA MINING TECHNIQUES Abstract Z.Y. Dong X. Li Z. Xu K. L. Teo School of Information Technology and Electrical Engineering The University of Queensland, St. Lucia, QLD 4072, Australia With the presence of competitive electricity market, accurate load and price forecasting have become essential for both system operator as well as general participants. Presented in this paper is an adaptive neuro-wavelet model for Short Term Electricity Load Forecast (STLF). Both historical load and temperature data, which have important impacts on load level, are used in forecasting by the proposed model. To enhance the forecasting accuracy by neural networks, the Non-decimated Wavelet Transform (NWT) is introduced to pre-process these data. The proposed model has been evaluated using actual data of electricity load and temperature of Queensland, Australia. The simulation results demonstrate that the model is capable of providing a reasonable forecasting accuracy in STLF. In addition, a forecasting scheme based on the presented model and data mining technique has been outlined for future research, which aims at forecasting electricity price with outliers. 1. INTRODUCTION The deregulation of power industries in Australia was recommended to the Federal and State Governments in 1991 [1]. Since then, competitive electricity markets, including wholesale and retail sectors, have been slowly but firmly introduced into serval states, resulting in formal launch of Australian Nation Electricity Market (NEM) in 1998. The fundamental objective of the deregulation is to increase the efficiency of electricity generation and distribution while maintaining sufficient security of operation. With the presence of electricity market, forecasting electricity demand and price become a necessity. Proper forecasting of electricity load would ensure adequate electricity generation to meet the consumers demands in the near and far future. In a deregulated electricity market, the penalty for inadequate load supply is very high. A generation company may lose its entire year s revenue due to unforeseen loss of generation caused by contingencies [2], usually out of system operation s control. One solution to overcome and to manage such contingencies is by proper system operational planning based on STLF, which forecasts the load of a few minutes, hours, or days ahead. The aim of STLF is to predict the future electricity demand based on the recognition of similar repeating trends of patterns from historical load data. However, other important factors, such as social events and especially the weatherrelated variables, must be considered [3], [4], [5], [6] and [7]. Traditionally forecasting models are mostly linear methods, such as Autoregressive Moving Average (ARMA) model, which have limited abilities to capture the nonlinearities in the time series such as electricity demand. In recent years, modern techniques based on artificial intelligence have shown promising results. The Neural Network (NN) based methods have gained most attention [8]. The NN is regarded as an effective approach and is now being used for electricity demand and price forecast. The reason of its popularity exists in its ease of

use and ability of function approximation of high complexity. Among different types of NNs that have been implemented, the Multilayer Percetrons (MLP) or Feed-forward NN has been proven to give satisfactory results for time series predictions [8] and [9]. In this paper a forecasting scheme has been developed. In addition to the historical electricity demand data, the weather (temperature), which has significant impacts on energy consumption, has also been used in the proposed method. By pre-processing these data using Non-decimated Wavelet Transform (NWT), the forecasting accuracy has been enhanced. An example using realworld data from Australia NEM has demonstrated the proposed method can provide reasonable accuracy in STLF. Furthermore, future scope of applying the scheme with data mining techniques in forecasting electricity spot price has been discussed. 2. THE PROPOSED WEATHER DEPENDANT SHORT TERM LOAD FORECASTING MODEL The proposed model has three stages for STLF, including data pre-processing using NWT, NN forecasting using feed-forward NN and data post-processing by NWT. Figure 1 shows the overall structure of the proposed model. 2.1 Data pre-processing Both the historical data of electricity load and temperature are treated as time series signals. NWT is used as the pre-signal processor in the proposed model. The à trous transform is an example of nondecimated wavelet transform. This algorithm is used in our approach to extract the hidden patterns of both the historical time series of electricity demand and temperature. If we consider a given time series signal, the à trous wavelet transform filters the signal through a series of low and high pass filters. The result from each filtering stage is the approximation (low frequency information) and detail (high frequency information) coefficient series obtained at different resolution levels. The number of filter stages depends on the highest resolution level determined for the filtering process. Comparing to classical wavelet algorithm, the NWT does not drop any data (non-decimation) and keep the time-invariance in decomposition, therefore has been adopted in the model. Figure 1 The proposed forecast model 2.2 Neural network forecasting In this stage, the wavelet coefficients obtained from NWT decomposition are fed into neural networks to predict future data at

one more time steps ahead. A set of feedfroward NNs are allocated to forecast the wavelet at different resolution levels. These networks contain only one hidden layer, which is adequate to approximate functions of any complexities [10]. The Scaled Conjugate Gradient algorithm (SCG) is used in training the NNs, due to its advantage of fast computational time for a large NN size. Figure 2 Wavelet Recombination Process 2.3 Data post-processing For post signal processing, the same wavelet technique and resolution level as mentioned in Stage 1 is used. In this stage, the outputs from the signal predictors (NNs) are combined to form the final predicted output. This is achieved by summing all the predicted wavelet coefficients. Figure 2 illustrates the recombination process. 3. SIMULATION AND RESULTS 3.1 Results The proposed model is tested with eight sets of historical data containing the electricity load and temperature data for the month of January 2001; one set of electricity demand data of Queensland Market and seven sets of temperature data from seven different locations (higher power consumption regions) of Queensland. It should be noted that the load data is with half-hour basis. The reason for using seven sets of temperature data is a compromise due to the unavailability of a general set of average temperature data for Queensland. The forecasting performance is measured by the mean absolute percentage error (MAPE) N 1 xi yi MAPE = *100% (1) N i= 1 xi where N is the number of points measured, x i is the actual values and y i is the predicted values. The model is trained with one week (336 points) of combined demand and temperature data for fifty cycles. The performance of the forecast model was evaluated and the results are as shown in Figures 3 to 5. Table 1: Summary of MAPEs No. of forecasted points MAPE (%) 336 (7 days) 1.4134 384 (8 days) 1.6198 432 (9 days) 2.0048 The MAPE values were calculated based on the forecasted values and the original time series. The MAPEs are recorded in Table 1.

Figure 3. 7-day ahead forecasting (336 points) Figure 5. 9-day ahead forecasting (432 points) 3.2 Discussion Figure 4. 8-day ahead forecasting (384 points) From the simulation results, it can be seen that the proposed model produces a reasonable accuracy in STLF. In our case study, training of the proposed model was conducted with only 336 points (one week) of the recent temperature and electricity historical data. Once the model is trained, it can be used to forecast the electricity load data for 48 times, each time forecasting one point ahead and with MAPE less than 1.7%, without the need to be retrained. Figure 6. The proposed model for multi-step forecasting Nevertheless, there is a limitation with the forecasting schemes that can be used for the model. The model can only predict one point ahead, which makes it unfeasible for the STLF of Queensland. To resolve the limitation, the model can be modified to

predict more points ahead. This modification can be done by including more prediction units into the model, which are used to predict the temperatures. The proposed model for multi-step prediction is shown in Figure 6. 4. DATA MINING APPROACH TO ELECTRICITY PRICE FORECASTING From the simulation results, it can be seen that the proposed model produces a reasonable accuracy in STLF. Once the model is trained, it can be used to forecast the electricity load data of future. However, the model can only predict one point ahead, which limits its application. To resolve the limitation, the model can be modified to predict more points ahead in future. Comparing to load data, forecasting the fairly chaotic electricity price is more complex. The high volatility of electricity price is attributed to the many exclusive features of the electricity and the power systems. Especially, the outliers or spikes in the price series have made the forecasting extremely difficult, although market participants rely much on this information for finical management. With the model proposed in this paper, we have further proposed an electricity forecasting scheme. Basic forecasting function will be achieved using the model developed in this paper, while the data mining technique is employed to account for outlier prediction. From the data set (see Figure 7), it can be seen that the total electricity demands (TED) are more or less periodical. The price spikes in RRP (regional average price) are correlated to the peak time of TED. As a result the price forecasting can be constrained by employing data mining for prediction of the time of the price spikes by correlation with the demand series in the following steps: 7000 6000 5000 4000 3000 TOTALDEMAND RRP 2000 1000 0 1 15 29 43 57 71 85 99 113 127 141 155 169 183 197 211 225 239 253 Figure 7. The correlation between total demands and price (RRP) 1. Calculate the correlation coefficient r (Pearson s r) between the Total Demand and RRP: r = ( x ( x i i X X ) ( yi Y ) 2 2 ) ( y Y ) i (2) sub-series s in the time period the t-w, to search D for a match s that satisfies the minimum pre-defined Euclidean distance d between s and s. 3. For s the RRP(Δt) * r will be the predicted price value. 2. For a given time series database D, at time t, with the window size w, we take a

Price series Demand series Weather series Preprocessing Base Series Figure 8 Proposed price forecasting scheme with outlier detection Combining the data mining technique with the forecasting model developed, the price forecasting scheme with outliers detection is illustrated in Figure 8. With the proposed outline, further research on electricity price forecasting with appearance of outliers is underway. 4. CONCLUSIONS Price Spikes This paper proposed a STLF model with a high forecasting accuracy. The NWT has been successfully implemented in the model. The implementation of NWT has reasonably enhanced the learning capability of the NNs in the model, thus minimizing their training cycles as shown in the simulations. Temperature has a close relationship with electricity load and its feasibility to be included for STLF has been proven with reasonable accuracy achieved from the simulations. In summary, the inclusion of temperature data (as an additional input variable) and the use of NWT (as the data processing tool) for the proposed STLF model have been proved to provide enhanced accuracy in STLF. The NN based forecast model can be further explored by combining data mining techniques in forecasting of electricity price signals, mainly price peaks based on the covariance between load and price data series of an electricity market. 5. REFERENCES [1] National Electricity Market Management Co Ltd (NEMMCO), Statement of Opportunities 2002. Correlation Processing Module Base Series NN- Wavelet Forecast Module Final Processing Final forecasted price signal [2] F. F. Wu and P. Varaiya, Coordinated multilateral trades for electric power networks: theory and implementation 1, Electrical Power and Energy Systems 21 (1999) 75-102. [3] G. Chicco et al, Load pattern clustering for short-term load forecasting of anomalous days, Proc. IEEE Porto PowerTech 2001, Sep. 2001. [4] D. W. Bunn, Forecasting loads and prices in competitive power markets, Proc. of the IEE, vol. 88, no. 2, Feb. 2000, pp. 163-169. [5] B. L. Zhang and Z. Y. Dong, An adaptive neural-wavelet model for short term load forecasting, Electric Power Systems Research 59 (2001) 121-129. [6] D. K. Ranaweera et al, Effect of probabilistic inputs on neural networkbased electric load forecasting, IEEE Trans. on Neural Networks, vol. 7, no. 6, Nov. 1996, pp. 1528-1532. [7] A. D. Papalexopoulos and T. C. Hesterberg, A regression-based approach to short-term system load forecasting, IEEE Trans. on Power Systems, vol. 5, no. 4, Nov. 1990, pp. 1535-1547. [8] H. S. Hippert et al, Neural networks for short-term load forecasting: A Review and Evaluation, IEEE Trans. on Power Systems, vol. 16, no. 1, Feb. 2001, pp. 44-54. [9] W. R. Foster et al, Neural network forecasting of short, noisy time series, Computers Chem. Enging., vol.16, no.4, 1992, pp. 293-297. [10] S. Haykin, Neural Networks, Macmillan College Publishing Company, Inc. 1994.