Elsevier

Ad Hoc Networks

Volume 123, 1 December 2021, 102658
Ad Hoc Networks

Hybrid RNN-LSTM deep learning model applied to a fuzzy based wind turbine data uncertainty quantization method

https://doi.org/10.1016/j.adhoc.2021.102658Get rights and content

Abstract

In addition to the plenty of advantages that the penetration of wind turbine (WT) brings to the smart networks, uncertainty problems can be considered as an unavoidable phenomenon that requires to be addressed. The results of high uncertainty are able to lead to the instability of management plans and schedules that is able to lead to the serious issues for operators and users. The following case study tries to propose a novel reinforcement learning based hybrid-based quantification technique to capture the prediction fault into the output power of WT. The offered approach has been applied the hybrid recurrent neural network (RNN) and long-short term memory (LSTM) layout with the aim of learning the utmost efficient Spatio-temporal properties of WT's output power. Because of the wide complication of the information, a novel optimization approach according to the modified sine cos algorithm has been suggested to aid in further steady layout training. The possibility and efficacy of the model are evaluated via the test analysis of two datasets in wind lands of Australian.

Introduction

Because of the random nature of the wind and the wide variety of it, expecting a wind agent with a rated capacity to always operate is difficult. So estimating the output power of the WT for an authentic power distribution of agents is necessary [1], [2], [3]. The performance curve of the WT and characteristics of wind speed are the two significant factors that determine the WT's output power technically. The producers of wind turbines (WTs) have been provides several curves to aid optimally cite the wind agent according to the performance curve of it, which are determined via 3 components: rated speed, cut-out and cut-in speed [4], [5], [6]. The efficiency curve of the usual WT supplied via its producer, that is displayed in Fig. 1. However the developer might set the initial factor, and also another factor which is the characteristics of wind speed is not controllable. Indeed characteristics of wind speed have been specified via plenty environmental parameters like moisture, solar radiation, human activities, season, temperature and etc. [7], [8], [9], [10]. This is an important issue in the rapid development of these significant and efficient renewable energy resources. Consequently majority of studies are started with the aim of finding the utmost authentic models to predict wind speed and power correctly [11].

Ref. [12] offered a multi-scale deep wind power (WP) forecasting method according to the unit-to-unit mapping. The stacked automatic encoder model has been applied in the form of a second solution with the aim of finding the wind speed correlations in diverse locations. The efficiency of the models has been evaluated in two feasible wind data sets. The wind uncertainty amount injected into the network has been presented in the form of a novel cost term that needs careful quantification in ref [13]. Additionally a possible vector-based procedure for predicting WP at the time of considering costs of wind dispatch has been offered. In addition, a risk index has been applied to penalize non-compliance with plans planned via the wind agent. Ref. [14] combined the long-term historical information with the updated real-time wind to create an authentic predicting method. Firstly, the offline WP forecast information is produced and afterward this information is upgraded to the area. Ref. [15] offered a successful ranking prediction method with the aim of supporting penetration of WP in the network via optimal transmission. In order to amend the predicting accuracy, an information-driven solution for predicting many short-term WP errors has been offered. Ref. [16] expanded a vibrational decomposition method which has broken the sample of the WP into several nonlinear, linear, and noisy elements. In the following method, the artificial grid models the nonlinear components, and also the regression layout models the linear components.Several probability density functions model the noise components. Ref. [17] concentrates on many short-term predicted WP which is applied the temporal-resolution layout. Afterward, topographies of the online changes have been recorded in a hidden prediction error. This procedure is provided excellent facilities in order to investigate the wind features according to the back propagation grids. Ref. [18] offered a hybrid method applying local decomposition and support vector regression to predict WP over a short time horizon. The setting variables of the layout have been optimized based on the firefly optimization method via inspiring by evolving optimization. It is proven that the firefly algorithm has the capability of providing an authentic and robust search structure to find the utmost optimal amounts. In the role of an optimizer, the firefly algorithm has been made according to the charm of insect light in hot areas. Ref. [19] examined the utmost successful characteristics of WP such as fault, temperature, step angle and wind speed in a deep learning (DL) procedure to provide suitable forecast outcomes. The simulation results support the effectiveness of DL against wind information samples. Ref. [20] expanded a multi-step approach to evaluate the great WP prediction according to the genetic algorithm and complement predictor. A communication-based machine learning vector has been used to solve the issue of vast complication in the frame of the learning grid. Ref. [21] applied short-term memory (LSTM) in the role of an in-depth scheme for learning reliable WP and output power. The double-decomposition according to the corrective solution has been proposed to improve the forecast outcomes. Alike tasks on the basis of wavelet transform [22], hybrid back propagation and decomposition [23] and chaotic time series [24] have been examined in the researches.

What is able to be inferred from the brief review is which precise forecasting of WT's output power is a valuable work for the permanent support of WTs in electrical networks. With a suitable performance curve supplied via the producer, a highly precise forecasting model is able to contribute to the effective and safe operation of renewable energy-based power networks. The following paper presents a new reinforcement learning based structure for potential WT's power prediction. For this purpose, the LSTM model has been composed with the concept of recurring network (RNN) [25] to model dynamic time behavior. In addition, a linear regression method has been offered to calculate the forecasted output power of the WT. It has been proven that because of randomness in wind models, there is a certain prediction error all the time. Consequently, the concept of point prediction is no longer a safe solution. The concept of predicting distance has the capability of being a great tool for managing the predict fault. The lower upper bound estimation (LUBE) with possible forecast indicators is the main solution for that transition. A new optimization procedure according to sine cos algorithm (SCA) has been suggested with the aim of adjusting the model setting parameters. Additionally, several modifications have been offered to raise the quest ability of the technique via enhancing the demographic variety. In summary, the main conclusion of the paper summarizes in the following:

  • Offering a probable reinforcement machine learning based forecasting method according to the hybrid RNN and LSTM.

  • Sketching forecasting fault via optimum forecasting interval.

  • Create a novel improved SCA (MCSA) to optimally adjust the parameters of the DL scheme.

  • Evaluate the efficiency of real wind experimental information sets in highly volatile winding areas.

The capabilities and characteristics of the model have been evaluated via the set of actual test data of a WTs in Australia. The basic objectives of the following paper are summarized in below; the hybrid DL model according to the LSTM and RNN is examined in part II. The MSCA in the role of the optimization tool is offered in part III. Part IV examines the probable forecasting method according to the LUBE. Part V explains the simulation results. Eventually, the conclusions of the paper is provided in part VI.

Section snippets

Hybrid reinforcement learning according to RNN and LSTM scheme

In the following part, a reinforcement based DL scheme according to the RNN and LSTM has been suggested to accurately predict the output power of WT. Technically, reinforcement learning (RL) is a kind of machine learning approach which allows an agent to train in an interactive way using trial and error using response from its own actions and experiences. It is tried to create the upper and lower boundaries of the scheme using prediction interval. This suggested scheme uses piled LSTM layers to

Enhanced Sine Cos method

To set the parameters of the offered DL model, anyone is in the need of having a strong optimizer with the aim of escaping from the trap of local optimization. The following paper considers LSTM systems that layers are stacked in sequence. Great forward and backward mathematical factors for accurate and guaranteed operation and discovery are the main cause to choose SCA. Additionally, proper convergence of the procedure has been supplied via some random factors that are placed in the

Probabilistic forecast and interval according to the indexes

The election of a possible forecast for the WT's output power is a requirement because of the highly sophisticated nature and nonlinearity of the input power of the device. For this purpose, the forecast distance is an attractive opinion that has the capability of providing a very good estimate of the predict information and in addition controlling the uncertainty of the predict error in wind samples. A forecast distance has been determined via its bandwidth and coverage level, that its quality

Experimental results

The following part concentrates on the simulation results on 2 WTs’ information set that are recorded in Australia. The instance recording frequency is each 15 min. The CAI membership function of fuzzy owns specifications for low and high amounts of 80 and 100, respectively. It helps to ignor each predicted distance with CAI less than 80%. In addition, the upper and lower limits of ABI are 40 and 20, respectively. To compare with further popular processes, just wind information recording is

Conclusion

Efficiency curves and characteristics of wind speed are 2 important ratios/indexes in creating a secure and reliable power production via the WT. Owning a performance curve via the producer, precise forecasting of wind speed/power can be considered as valuable work. The following study presents a new possible plan based on reinforcement learning and DL scheme for precise forecasting and sketching of the WT instance information. For this purpose, the combined method of RNN and LSTM with LUBE can

Declaration of Competing Interest

None.

Lihua Lin, PhD, senior engineer, master tutor. She has presided over and participated in 9 related scientific research projects in the research field, with a research funding of nearly 6 million yuan. She has published more than 20 papers in domestic core journals, including 5 papers retrieved by the first author EI, and 2 co-edited works. Authorized 3 national patents and 1 software copyright. Won two science and technology awards from Xi'an University of Science and Technology. The main

References (29)

  • M. Dabbaghjamanesh et al.

    Effective scheduling of reconfigurable microgrids with dynamic thermal line rating

    IEEE Trans. Ind. Electron.

    (2018)
  • N. Moustafa et al.

    Generalized outlier gaussian mixture technique based on automated association features for simulating and detecting web application attacks

    IEEE Trans. Sustain. Comput.

    (2018)
  • M. Dabbaghjamanesh et al.

    A novel distributed cloud-fog based framework for energy management of networked microgrids

    IEEE Trans. Power Syst.

    (2020)
  • M. Etemadi et al.

    Wind turbine output power prediction in a probabilistic framework based on fuzzy intervals

    Iran. J. Sci. Technol. Trans. Electr. Eng.

    (2021)
  • Cited by (10)

    • An introduction of a reward-based time-series forecasting model and its application in predicting the dynamic and complicated behavior of the Earth rotation (Delta-T values)[Formula presented]

      2021, Applied Soft Computing
      Citation Excerpt :

      Therefore, they become highly complex in terms of the learning period, especially for long time periods. One special type of RNNs is long-short term memory (LSTM), which is capable of learning both short-term and long-term dependencies in data [31,32]. The weight updates and preferred optimization methods of LSTM are the same as those of RNN [32].

    • SHORT-TERM WIND SPEED FORECAST BASED ON WRF SIMULATION AND ATTENTION MECHANISM

      2023, Taiyangneng Xuebao/Acta Energiae Solaris Sinica
    • Reinforcement learning in wind energy - a review

      2023, International Journal of Green Energy
    View all citing articles on Scopus

    Lihua Lin, PhD, senior engineer, master tutor. She has presided over and participated in 9 related scientific research projects in the research field, with a research funding of nearly 6 million yuan. She has published more than 20 papers in domestic core journals, including 5 papers retrieved by the first author EI, and 2 co-edited works. Authorized 3 national patents and 1 software copyright. Won two science and technology awards from Xi'an University of Science and Technology. The main research areas are information systems and informatization, distribution automation, power system safe operation, power equipment safety monitoring, fault diagnosis, and early warning.

    Min Li was born in Shandong Province, SD, CHN in 1995, majoring in electronics and communications. The first-year graduate student won the first-class scholarship, familiar with machine learning and deep learning, participated in the training of senior big data analysts, and has reached the special technical level; mastered various development languages such as Python, Matlab, and c language, and participated in national key research and development projects "Soul Core" DSP application function library and software middleware development, the project has completed acceptance. Department of Telecommunication and information, Xi'an University of Science and Technology, 118249 Xi'an China 710054

    Li Ma was born in Lanzhou, Gansu Province in 1979. She received the B.S. in electronic information engineering from Xi'an University of Science and Technology, Shaanxi, China, in 2002, and the M.S. degree in information and communication engineering from Xi'an University of Science and Technology, Shaanxi, China, in 2005, and the Ph.D. in safety science and technology from Xi'an University of Science and Technology in 2014. From 2008 to 2014, she was a Lecturer with the Department of Internet of Things Engineering, Department of Telecommunication and information, Xi'an University of Science and Technology. Since 2014, she has been an associate professor with the Department of Internet of Things Engineering, School of Communication, Xi'an University of Science and Technology. She is the author of one book, more than 15 articles. Her research interests include data mining, big data analysis, Internet of Things technology, and intelligent applications.

    Aliasghar Baziar received the M.S.E.E. degree from Dezfool Azad University. He is currently pursuing the Ph.D. degree in power engineering. He has six year experience of working as a senior engineer in different parts of the industry and five year experience as a faculty member of Azad University. His research interests include power system operation and management, optimization in power systems, renewable energy sources, and uncertainty modeling. Department of Telecommunication and information, Sarvestan University, Sarvestan, Iran

    Ziad M. Ali received the B.Sc. degree in electrical engineering from Assiut University, faculty of engineering, Assuit, Egypt, in 1998. He worked as a demonstrator in Aswan faculty of engineering, south valley university, Aswan, Egypt. He obtained an M.Sc. degree from Assiut University, faculty of engineering in electrical engineering in 2003. He worked as Assistant Lecturer in Aswan faculty of Engineering. He obtained a Ph.D. degree in 2010 from Kazan State Technical University, Tatarstan, Russia. He is currently working as Associate Professor in Electrical Dept. College of Engineering at Wadi Addawasir, Prince Sattam bin Abdulaziz University, Saudi Arabia. College of Engineering at Wadi Addawaser, Prince Sattam Bin Abdulaziz University, KSA, 11991 Electrical Engineering Dept., Faculty of Engineering, Aswan University, Egypt, 81542

    View full text