- Original article
- Open Access
RBF neural network-based online intelligent management of a battery energy storage system for stand-alone microgrids
© Kerdphol et al. 2016
- Received: 1 October 2015
- Accepted: 3 February 2016
- Published: 23 February 2016
An offline optimization approach based on energy storage management response in a microgrid was not fast and not reliable enough to control and adjust the system efficiently after the loss of the utility grid. Thus, it causes system inefficiency and collapse in the presence of violent changes of loads or outage of distributed generations. To solve such a problem, more real-time management is needed. Any changes in loads/generations should be compensated successfully by a battery energy storage system (BESS) in a short period of time.
This paper presents a new method for the intelligent online management of both active and reactive power of a BESS based on a radial basis function neural network (RBFNN) incorporating particle swarm optimization (PSO) to prevent the stand-alone microgrid from instability and system collapse. BESS is centrally controlled by a controller developed by the proposed RBFNN. PSO is used to determine the optimized active and reactive power at every load/generation changing situation to monitor the effect of system frequency, voltage, and reference power regulation. These optimized power data are then employed as target data for the RBFNN generalization and training process. To enable the online updating of the operating parameters, the proposed RBFNN is implemented in the management process. With an appropriate RBFNN training, the optimum active and reactive power can directly be obtained without the necessity of performing the PSO optimization process at any change of load/generation.
The results show that the predictive results of the proposed RBFNN model only slightly differed from the target results based on PSO and have a minimum statistical error compared to the predictive results based on the multilayer perceptron neural network (MLPNN) model.
The proposed RBFNN is suitable for the online estimation of the active and reactive power of BESS and can be used for real-time energy storage management as an online controller.
- RBF neural network
- Battery energy storage system
- Energy management
- Frequency and voltage control
During the great earthquake and tsunami in March 2011 and the heavy snowfall disaster in Tokushima, Japan, in December 2014, thousands of people had no access to electricity. IP telephone systems using the Internet which spreads in almost all families were not able to operate at that time. It caused difficulties to rescue teams for providing assistance. Moreover, victims who lived in their homes could not get warm as their heaters were not able to operate at the time of the disaster . To solve such a situation, facilities are needed to provide electricity at the time of a blackout or disaster. Battery energy storage systems (BESS) can offer a good solution to such a system. Advantages of BESS include an improvement of the system frequency, especially when BESS is used for system frequency control. For small disturbances, BESS is discharging when the system frequency is lower than the nominal frequency of 50 or 60 Hz. On the other hand, BESS is charging when the system frequency is higher than the nominal frequency of 50 or 60 Hz. For large disturbances, BESS can enhance the performance of the system frequency control by integrating BESS with an under-frequency load shedding scheme or an under- or over-frequency generation trip. With these different functions, it can be concluded that BESS is a rapid and flexible element for power systems [2–4].
After the 2011 Japan earthquake and tsunami in the Tōhoku area, a micro/smart grid developed in Japan has been focused on resilience. As a solution of micro/smart grids, it promised to simplify the wide penetration of renewable energy sources (RESs) and BESS units into the power system and increase the reliability of electrical supply to consumers, but decrease system losses and greenhouse gas emissions. Due to their potential benefits of providing secure, reliable, efficient, sustainable, and environmentally friendly electricity from RESs, micro/smart grids have received great attention and became remarkable in electricity .
A concept of a micro/smart grid is demonstrated as a system that can intelligently integrate the actions of all users incorporating generators or loads in a manner suitable for providing an economically sustainable and secure power system . All signals at loads/generations will be processed by the system management and react to the situations which occurred optimally. By an intelligent management of the active and reactive power of BESS for a stand-alone microgrid, this technique can prevent the stand-alone microgrid from instability and collapse in the presence of violent changes of loads or outage of distributed generations. In a number of studies, the aspect of a managing reference power of distributed generations in the distribution system has been presented [7–9]. In , the management of BESS power for the typical 4Q-load has been proposed and analyzed. Furthermore, an offline optimization approach based on a real-time energy storage management was proposed in . To control the system as efficiently as possible, more real-time management is needed. Any changes in loads/generations should be compensated successfully by the BESS in a short period of time. To perform the real-time online management operation in this study, radial basis function neural networks (RBFNNs) seem to be most suitable for such an online modeling method in terms of a fast time calculation process and instant responses. The advantages of RBFNN are two major issues: the training processes are substantially faster than the multilayer perceptron neural networks (MLPNN) and RBFNN does not encounter with the local minima problems [12, 13]. RBFNN provides a very significant tool for optimization tasks as they are extremely powerful computational devices with the capability of parallel processing, learning, generalization, and universal approximation [14, 15].
This paper deals with the online intelligent management of active and reactive power of the BESS installed in the microgrid to prevent the microgrid from instability and collapse in the presence of violent changes of loads or outage of distributed generations after the loss of the utility grid (e.g., blackout or disaster). The BESS is centrally controlled by a controller developed by the proposed RBFNN. The active and reactive power of the BESS are managed by using the RBFNN incorporating a particle swarm optimization (PSO) process with the objective of maintaining the frequency and voltage of the stand-alone microgrid within acceptable ranges. First, the optimum settings of the BESS which are the optimized active and reactive power are determined by PSO. The PSO process has to reply to every change in load/generation to achieve the optimum operating conditions for the entire system. These optimized operating data are then applied as target data for RBFNN generalization and training processes. To enable the online updating of the operating parameters, the proposed RBFNN is implemented in the management process, and the database extracted from the PSO process is used as target data in the RBFNN generalization and training. With an appropriate RBFNN training, the well-trained RBFNN can be employed as the online mode where the system is using new input data. By applying the proposed RBFNN approach in the system, the optimum active and reactive power of the BESS can directly be obtained without the necessity of performing the optimization process-based PSO at any change in load/generation. The predictive results of the proposed RBFNN are compared with the predictive results of the MLPNN, and it is clearly shown that the proposed RBFNN-based online management method gave the best performance in predicting the optimum active and reactive power of the BESS for the microgrid.
This paper, compared to other previous research contributions, deals with the ability to control the system as efficiently as possible by designing and implementing the real-time/online management of the BESS based on the proposed RBFNN for using it as an online controller to control, manage, and prevent the microgrid from instability and collapse in the presence of violent changes of loads or outage of distributed generations after the loss of the utility grid.
Microgrid study system
Solar photovoltaic generation
RESs are depending on weather conditions. Thus, a BESS is used to store surplus electrical energy to maintain the system frequency and voltage and supplies the power for loading into a microgrid in the case of low solar ratio or load changes. Moreover, the BESS can smooth the fluctuation of solar radiation and enhance the load availability. A more detailed BESS information, along with the most BESS models, is presented in various research papers, for instance in [17, 18].
In this study, in the case of a grid-connected mode, in which the power generated by the microgrid system is higher than the load demand, the surplus power can be stored in a BESS for future uses. On the contrary, when there is any shortage in the power generation of the microgrid, the stored power is used to supply the load. For a stand-alone mode, the main purpose of the BESS is to stabilize the microgrid from instability and collapse in the presence of violent changes of loads or outage of distributed generations [3, 4].
By performing the offline optimization process, the optimized active and reactive power set points of the BESS are obtained. These optimized data are then used as target data for RBFNN/MLPNN generalization and the training process in the online optimization. With an appropriate RBFNN/MLPNN training process, the well-trained RBFNN/MLPNN can be used as the online mode where the system is using new input data and then the optimum active and reactive power of the BESS will be obtained and remotely adjusted via the communication link to the BESS.
Test system explanation
The proposed online management approach is applied to the typical microgrid system illustrated in Fig. 1. To achieve a smart grid system with online capability, the microgrid is integrated in a highly developed communication technology which is also connected with the BESS as well as distributed generations and loads, thus enabling the coordinated control between generations and loads. The microgrid will be scanned by a data acquisition device for every half an hour of load demand for a 24-h operation. The data acquisition device measures the power output of solar PV and all load profiles. The acquired information is used to control the optimum operation of the BESS in subsequent intervals. The optimum active and reactive power of the BESS will be calculated by the control system so that they can be remotely adjusted via the communication link instantly after the loss of the utility grid.
ANN-based online optimum active and reactive power of the BESS
PSO-based offline optimization (stage 1)
The PSO process has to reply to every change in the load/PV generation in order to achieve the optimum operating conditions for the overall system. Hence, after any change of the load/PV data, the optimized power setting of the BESS will also be changed, with the result that a new optimization process is needed.
The learning factors have important effects on the algorithm convergence rate. Future information for PSO can be found in [20–23]. In this study, the number of iteration is 30. Learning factors are c 1 and c 2 which are equal to 1.4940. The inertia weight is 0.7920.
Artificial neural network-based online optimization (stage 2)
Artificial neural network (ANN) is simulating the brain of humans in processing information through a series of interconnected neurons. It is one of the famous prediction models as it has the remarkable ability of mapping complex and highly non-linear input-output patterns without the knowledge of the actual model structure. The RBFNN and the MLPNN are widely used in ANN structures, and their roles affect the network performance. Nowadays, a performance comparison of the RBFNN and the MLPNN in several applications has attracted the attention of researches [24–26]. Nevertheless, no comparison has been carried out so far between the intelligent management of active and reactive power of the BESS for a microgrid management system. Hence, this work proposed the RBFNN-based online management of the BESS and selected the MLPNN-based online management of the BESS for the performance comparison in the system. To enable the online updating of the operating parameters, the RBFNN/MLPNN is proposed and implemented in the management process for the second stage. Afterwards, the output performance of the RBFNN and the MLPNN were investigated and compared in the ANN training and testing results. During this stage, the database extracted from the PSO process was used as a target data in the RBFNN/MLPNN generalization and training process.
RBFNN-based online management of the BESS
The RBFNN is a type of feedforward neural network which learns using a supervised training method. Radial functions are a special class of functions, and their characteristic feature is that the response decreases or increases with the distance from a center point. It is obvious that the RBFNN is able to approximate any reasonable continuous function mapping with a satisfactory level of accuracy [13–15]. In this paper, the proposed RBFNN consists of three layers of neurons: an input layer, a hidden layer, and an output layer.
MLPNN-based online management of the BESS
The MLPNN belongs to the class of feedforward networks. In the MLPNN structure, this structure is established in a layered feedforward network and is contained by an input layer, one or more hidden layers, and an output layer. The weight total of the input data and the chosen bias are passed through a transfer function to obtain the output data. The number of hidden layers is able to be changed based on the problem data in the training process [12, 27]. In this paper, the MLPNN consists of three layers of neurons, which demonstrates that only one hidden layer is included, and one type of activation function is used in the hidden layer and one output layer is contained.
Artificial neural network parameters
Hidden layer neurons
Output layer neurons
Step 1. Obtain input data and target from the PSO process.
Step 2. Create the RBFNN/MLPNN network and train the network until the conditions of the network setting parameters are reached.
Step 3. Test the network and control the regression analysis.
Step 4. Store the trained network. Afterwards, the trained network is ready to be tested by using new input data for this online process. Please consider that step 1 to step 4 belong to the offline process.
Step 5. Process new input data to the online process and obtain the optimum power data of the BESS.
Results and discussion
This part describes the results of the proposed online predictive power management of the BESS with the objective of frequency and voltage control of the stand-alone microgrid by using the proposed RBFNN. Thus, the proposed RBFNN approach will automatically determine the optimum power of the BESS in order to prevent the isolated microgrid from instability and collapse in the presence of violent changes of loads or outage of distributed generations. Afterwards, the predictive results of the proposed RBFNN model are compared with the MLPNN model, considering the error efficiency and positional accuracy.
ANN training results
There are 12 inputs being the time step of the five load demands and the solar PV data which were fed into the RBFNN/MLPNN controller. The outputs of the neural network will be the predictive results of the optimum active and reactive power of the BESS which will determine the optimum operation of the BESS for the stand-alone microgrid.
After the inputs and targets for the training data are initiated, the next process is the separation of the data for training, validation, and test. During this stage, 70 % of the sample data are used for the training process (i.e., 940 data), 15 % of the sample data are used for validation (i.e., 202 data), and 15 % for the test data (i.e., 202 data).
ANN testing results
Error indexes for online optimum BESS management
Y 1-based RBF (active power)
9.947 × 10−7
Y 2-based RBF (reactive power)
7.821 × 10−7
Y 1-based MLP (active power)
Y 2-based MLP (reactive power)
Frequency and voltage of stand-alone microgrids
In this paper, a novel method for the optimum online intelligent management of active and reactive power of the BESS for the isolated microgrid is proposed. The entire BESS is centrally controlled by a controller developed using the proposed RBFNN model. The results show that the proposed RBFNN is able to follow the optimum target-based PSO at almost every time step mentioned under the changes of typical loads and solar PV generation with the profiles of a sunny and a rainy day. Compared with the MLPNN model, the proposed RBFNN model provides superior performance, when the error efficiency and positional accuracy are considered. It can be summarized that the proposed RBFNN model is appropriate for the online prediction of active and reactive power of the BESS differing only slightly from the optimal target result-based PSO and can consequently be used for real-time energy storage management as an online controller.
The authors are grateful for the feedback of two anonymous reviewers and the editors of this journal; their comments helped us a lot in improving the quality of this paper. We would like to thank Dagmar Fiedler, Editorial Manager for the revisions.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
- Marnay C, Aki H, Hirose K, Kwasinski A, Ogura S, Shinji T (2015) How two microgrid fared after the 2011 earthquake. IEEE Power Energy Magazine 13(3):44–57View ArticleGoogle Scholar
- Wu D, Tang F, Dragicevic T, Vasquez J, Guerrero J (2014) Autonomous active power control for islanded AC microgrid with photovoltaic generation and energy storage system, IEEE Trans. Energy Conversion 29(4):882–892View ArticleGoogle Scholar
- Moghadam MRV, Zhang R, Ma RTB (2013) Randomized response electric vehicles for distributed frequency control in smart grid. In: proceeding of IEEE International conf. on Smart Grid Communications., pp 139–144Google Scholar
- Nan Z, Qian L, Li H (2014) Auxiliary frequency and voltage regulation in microgrid via intelligent electric vehicle charging. In: proceeding of IEEE International conf. on Smart Grid Communications., pp 662–667Google Scholar
- Inthamoussou F, Pegueroles-Queralt J, Bianchi F (2013) Control of a supercapacitor energy storage system for microgrid application, IEEE Trans. Power Systems 28(3):690–697Google Scholar
- Cecati C, Mokryani G, Piccolo A, Siano P (2010) An overview on the smart grid concept. In: Proceeding of Conf. on 36th annual IEEE Industrial Electronics Society., pp 3322–3327Google Scholar
- Toma S, Senjyu T, Miyazato Y, Yona A (2008) Decentralized voltage control in distribution system using neural network. In: proceeding of Conf. on 2nd IEEE International Conference., pp 1557–1562Google Scholar
- Hatta H, Kobayashi H (2007) A study of centralized voltage control method for distribution system with distributed generation. In: proceeding of Conf. on 19th CIRED International Conference on Electricity Distribution., pp 1–4Google Scholar
- Marcela M, Sumper A, Gomis O, Sudra A (2011) Reactive power dispatch in wind farms using particle swarm optimization technique and feasible solutions search. Appl Energy 1:4678–4686Google Scholar
- Baalbergen F, Baber P, Ferreira J (2009) Energy storage and power management for typical 4Q-load. IEEE Trans on Industrial Electronic 56(5):1485–1498View ArticleGoogle Scholar
- Rahbar K, Xu J, Zhang R (2015) Real-time energy storage management for renewable integration in microgrid: an off-line optimization approach. IEEE Trans on Smart Grid 6(1):124–134View ArticleGoogle Scholar
- Li G, Na J, Stoten D, Ren X (2014) Adaptive neural network feedforward control for dynamically substructured systems, IEEE Trans. Control System Technology 22(3):944–954View ArticleGoogle Scholar
- Bianchini M, Frasconi P, Gori M (1995) Learning without local minima in radial basis function networks, IEEE Trans. Neural Netw 6(1):749–756View ArticleGoogle Scholar
- Park J, Sandberg IW (1991) Universal approximation using radial basis functions network. Neural Comput 3(2):246–257View ArticleGoogle Scholar
- Tang Y, Wong WK (2015) Distributed synchronization of coupled neural networks via randomly occurring control. IEEE Trans on Neural Networks and Learning Systems 24(3):435–447View ArticleGoogle Scholar
- Sitbon M, Leppaho J, Suntio T, Kuperman A (2014) Dynamic of photovoltaic-generator-interfacing voltage-controlled buck power stage. IEEE Journal of Photovoltaics 5(2):633–640View ArticleGoogle Scholar
- Kottick D, Blau M, Edelstein D (1993) Battery energy storage for frequency regulation in an island power system, IEEE Trans. Energy Convers 8(3):455–458View ArticleGoogle Scholar
- Kerdphol T, Qudaih Y, Mitani Y (2015) Optimal battery energy storage system size using particle swarm optimization for microgrid system. International Review of Electrical Engineering (IREE) 10(2):277–285View ArticleGoogle Scholar
- Aditya SK, Das D (1999) Application of battery energy storage system to load frequency control of an isolated power system. Int J Energy Res 23(1):247–258View ArticleGoogle Scholar
- Gaing ZL (2004) A particle swarm optimization approach for optimum design of PID controller in AVR system, IEEE Trans. Energy Conversion 19(2):284–391Google Scholar
- Esmin AA, Lambert-Torres G, Zambroni de Souza AC (2005) A hybrid particle swarm optimization applied to loss power minimization, IEEE Trans. Power Systems 20(2):859–866View ArticleGoogle Scholar
- Gaing ZL (2003) Particle swarm optimization to solving the economic dispatch considering the generator constraints, IEEE Trans. Power Systems 18(3):1187–1195View ArticleGoogle Scholar
- Kennedy J, Eberhart R (1995) Particle swarm optimization in Proc IEEE Int. Conf. Neural Netw 4:1942–1948Google Scholar
- Zyandehroodi H, Mohamed A, Shareef H, Mohammadjafari M (2010) Performance comparison of MLP and RBF neural networks for fault location in distribution networks with DGs, Proceeding of IEEE Int Conf Power Energy., pp 342–345Google Scholar
- Sereno F, de Sa JP M, Matos A, Bernades J (2000) A comparative study of MLP and RBF neural nets in the estimation of the foetal weight and length. In: Proceeding of RECPAD., pp 1–9Google Scholar
- Memarian H, Balasundram SK (2012) Comparison between multi-layer perceptron and radial basis function networks for sediment load estimation in a tropical watershed. Journal of Water Resource and Protection 1:870–876View ArticleGoogle Scholar
- Karami H, Mousavi MF, Shamsipur M, Riahi S (2006) New dry and wet Zn-polyaniline bipolar batteries and prediction of voltage and capacity by ANN. J Power Sources 154(1):298–307View ArticleGoogle Scholar