Multistage energy management system using autoregressive moving average and artificial neural network for day‐ahead peak shaving
2019; Institution of Engineering and Technology; Volume: 55; Issue: 15 Linguagem: Inglês
10.1049/el.2019.0890
ISSN1350-911X
Autores Tópico(s)Solar Radiation and Photovoltaics
ResumoElectronics LettersVolume 55, Issue 15 p. 853-855 Power electronics, energy conversion and sustainabilityFree Access Multistage energy management system using autoregressive moving average and artificial neural network for day-ahead peak shaving K. Mahmud, Corresponding Author K. Mahmud khizir.mahmud@unsw.edu.au orcid.org/0000-0002-9539-7033 School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, NSW, 2052 AustraliaSearch for more papers by this authorA. Sahoo, A. Sahoo School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, NSW, 2052 AustraliaSearch for more papers by this author K. Mahmud, Corresponding Author K. Mahmud khizir.mahmud@unsw.edu.au orcid.org/0000-0002-9539-7033 School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, NSW, 2052 AustraliaSearch for more papers by this authorA. Sahoo, A. Sahoo School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, NSW, 2052 AustraliaSearch for more papers by this author First published: 01 July 2019 https://doi.org/10.1049/el.2019.0890Citations: 8AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract In this Letter, a multistage energy management system is developed and analysed incorporating the coordination between the tertiary stage and the primary stage. In the tertiary level, a day-ahead peak shaving strategy is developed using the autoregressive moving average and artificial neural network technique. These two techniques are used to predict customer's power demand and photovoltaic power generation, which are fed to a tertiary controller for the day-ahead power demand management. As the peak-power-demand management system is highly dependent on the demand-generation values, any fluctuations and errors in predicted values impact the performance of the peak shaving. The reference power generated from the energy management layer at the tertiary stage is communicated to the local inverter controller at the primary stage. The inverter implements a dq-current controller to track the reference power efficiently. Introduction Renewable energy sources (RES) are highly intermittent, and their integration to the microgrid (MG) imposes additional control and management challenges [1]. A coordinated control strategy having a robust energy management system (EMS) is essential for reliable operation of MG. The EMS can predict either power generation considering the weather dynamics and intermittency of the RES, and coordinates between various layers of the hierarchically coordinated MG control [2]. In this Letter, two techniques, i.e. autoregressive moving average (ARMA) and artificial neural networks (ANNs) are modelled to predict the customer's power demand and photovoltaic (PV) power generations. A multistage EMS algorithm is developed to use these predicted values and shave the peaks in a day ahead of the actual operation. The prediction of load demand during different peak hours over a day decides the power generation reference which is communicated to the local inverter control. Energy management algorithm The proposed multistage EMS consists of a roof-top PV and battery storage, as shown in Fig. 1. Fig 1Open in figure viewerPowerPoint Multistage power and energy management architecture This battery storage and PV is connected to the main domestic bus through an intermediate DC bus. In the DC bus, PV and battery storage are connected using a unidirectional DC–DC and a bidirectional DC–DC converter. The DC bus is connected to the main bus using a bidirectional DC–AC/AC–DC converter. Let us assume the power drawn from the AC bus as a function of power and time is expressed as (1)The power demand in (1) is predicted using ARMA and ANN. The output from an ARMA-based prediction system is expressed as [3, 4] (2)where is the predicted output by the ARMA model and , and m and n are the order of the ARMA model, and are the model parameters, is the random error, and is a constant. On the other hand, ANN uses hidden layers between its inputs and outputs. If it consists of g number of hidden layers, and k number of inputs, for inputs, the output is given as [3, 4] (3)where are the weights from hidden layers, and are bias terms, and is a random shock. Let us assume the reference power demand is . The peak and off-peak periods are identified based on their relationship with the predicted power demand (4)For , the available power to charge the battery storage expressed as (5)where is the PV power output. Let us assume that the current state-of-charge of battery storage is (in per cent) and the maximum charging limit is (in per cent). If the capacity of the battery storage is . The required power to charge the battery storage from is expressed as (6)For , the required power to shave the peak is expressed as (7)If , the peak power demand will be supplied by the PV. However, if , the additional power will be supplied by the battery storage. If the lower discharging boundary of the battery storage is , the maximum power battery storage can provide is (8)Although the system requires power power to shave the peak, the maximum power the available resources can provide is . So, the new load curve after peak shaving will follow (9)The proposed day-ahead peak-shaving algorithm is described step by step in Algorithm 1. Algorithm 1.Day-ahead peak shaving strategy 1: Initialisation for ANN: I. Define the input–output functions, i.e. II. Fix the functional algorithm, i.e. Bayesian regularisation III. Set the number of hidden layers and train the network using Initialisation for ARMA: I. Set the parameters II. Set the values of m and n 2: Prediction: predict , , using ARMA and ANN 3: while 4: check: PV power generation 5: calculate: the available power check: battery charging constraints , , 6:while 7: check: PV power generation calculate: the required shaving power 8: if 9: 10: else if 11: check: check: battery discharging constraints , , 12: calculate: maximum power battery can support 13: 14: check steps 3 and 6 to continue the process} Inverter local controller The grid-connected PV system employs cascaded power and current control to feed the prerequisite amount of power to the grid. The outer control is power control. The dq-components of voltage and current which are estimated using a synchronous reference frame phase locked loop are used to calculate the three-phase active and reactive power (Pm and Qm) for inverter connected to the grid. The measured power is compared with the reference power (Pref) which is obtained from the supervisory controller explained in the above section. The error is fed to the proportional and integral controller to generate the reference current for both d and q axes () as shown in (10) and (11). Active power error is responsible for the d-component while the reactive power generates the q-component (10) (11) Case studies The consumer's power demand and local renewable energy generation, i.e. PV power, are important parameters for the battery storage control and provide load-support to customer. The proposed system is tested using real Australian power distribution system, which is in Elermore Vale, NSW. The ARMA and ANN techniques are applied to predict the power customer's power consumption pattern and their performance is shown in Fig. 2a. The real local weather data is fed to the PV cells and based on that the PV power output is predicted, as shown in Fig. 2b. Fig 2Open in figure viewerPowerPoint Power demand and PV power generation prediction a Power demand prediction using ANN and ARMA b PV power generation prediction using ANN and ARMA Proper tuning and training of the prediction model are required to get reliable performance. For example, in the case of ARMA, the tuning of its order as shown in (2) is required. The impact of ARMA order (m, n) to the prediction performance is shown in Fig. 3a. In case of uncertainty, the prediction error increases, as shown in Fig. 3b, and it impacts the performance of the EMS. Fig 3Open in figure viewerPowerPoint Prediction performance a Impact of tuning to prediction performance b Impact of prediction performance during uncertainty The ANN and ARMA predicted values are fed to the EMS for day-ahead management using the Algorithm 1, and (1)–(9). The performance of the EMS is shown in Fig. 4. The PV reference power processed in the tertiary layer is provided as reference power for an inverter local controller. The tracking of the inverter d-axis current and the three-phase inverter output current is shown in Figs. 5a and b. Fig 4Open in figure viewerPowerPoint Performance of day-ahead energy management systems Fig 5Open in figure viewerPowerPoint Performance of inverter local controller a Id current trace b Iabc current trace c Running THD for Iabc Performance evaluation of inverter local controller It is observed that the dq-current controller is operating in fine tune to track the reference power. The moving total harmonic distortion (THD) of the inverter output current is plotted in Fig. 5c. The THD level is found to lie in a range between −18.5 dB and −17.75 dB. Conclusion The Letter presents a day-ahead management system considering both tertiary and primary control layers. The result shows to be effective to flatten the consumer's load curve, hence reduce energy costs and improve the load factor. The performance of the dq-current controller is tested in terms of efficient current tracking and running THD calculations using the reference power generated from the tertiary stage. References 1Ma, Z., Pesaran, A., Gevorgian, V. et. al.,: 'Energy storage, renewable power generation, and the grid: NREL capabilities help to develop and test energy-storage technologies', Electr. Mag., 2015, 3, (3), pp. 30– 40 (https://doi/org/10.1109/MELE.2015.2447972) 2Stenclik, D., Denholm, P., Chalamala, B.: 'Maintaining balance: the increasing role of energy storage for renewable integration', Power Energy Mag., 2017, 15, (6), pp. 31– 39 (https://doi/org/10.1109/MPE.2017.2729098) 3Adhikari, R., Agrawal, R.K.: ' An introductory study on time series modeling and forecasting', arXiv Prepr. arXiv1302.6613, 2013 4Alanis, A.Y.: 'Electricity prices forecasting using artificial neural networks', Lat. Am. Trans., 2018, 16, (1), pp. 105– 111 (https://doi/org/10.1109/TLA.2018.8291461) Citing Literature Volume55, Issue15July 2019Pages 853-855 FiguresReferencesRelatedInformation
Referência(s)