Original articlesImproved butterfly optimization algorithm applied to prediction of combined cycle power plant
Introduction
The utilization of combined cycle power plants (CCPP) is expanding all over the world as the demand for electricity rises [33]. CCPP is a large power plant that combines a gas turbine and a steam turbine. They use fossil fuels to achieve the highest technical efficiency of power generation, which has now exceeded 0.60 [23].
The essential load operation of a power plant is influenced by four key elements that determine the electric power output: ambient temperature, atmospheric pressure, relative humidity, and exhaust pressure [36]. Any change in these parameters may change the power output of the system. Effectively predicting the change of output caused by the change of parameters significantly impacts power production. Only by precisely anticipating the full-load power production of base-load power plants can electricity consumption be better managed and allocated.
In the literature [13], by using the artificial neural network (ANN) model of the multi-layer feedforward network model, the operation and performance parameters of a thermal gas turbine in different environments are accurately predicted. The application of intake heating technology in CCPP is proposed in Ref. [38]. The results show that when the inlet gas temperature is higher, the thermal efficiency of the thermal gas turbine will be reduced, and the energy conversion efficiency of CCPP will be improved. The gas turbine cycle is analyzed and studied in Ref. [27]. A better rational efficiency value of the turbine is obtained by using the response surface method (RSM) to optimize the input variables. In Ref. [7], a novel sequence model is proposed that can detect the degree of engine component degradation, minimize the computational burden of diagnosis procedures, and increase gas turbine reliability and energy efficiency. In [14], [22], [31], ANN is used to predict the operation characteristics and efficiency of burners and turbines.
Previous researchers mainly built models for evaluation through machine learning and neural networks. SVR in machine learning has achieved remarkable results in data regression prediction. In recent years, the public has widely known intelligent optimization algorithms because of their excellent optimization ability, and it is applied to the parameter optimization of the model. ANN is one of them, and it excels at coping with non-linear situations. At present, we are in an era of big data, and there must be a tremendous amount of information changes in the development of everything. Combining these data analyses with existing models for analysis and prediction, can effectively avoid some unnecessary losses. In [39], the model is applied to the field of ocean exploration and early warning to predict the height of waves, and hybrid thinking evolutionary algorithm (MEA) paired with BP neural network (BPNN) is utilized to improve the prediction and generalization capacity of BPNN [26]. By using chest X-rays to detect COVID-19 and other pneumonia, a new neural network structure is presented to speed up the doctor’s differential diagnosis of patients. SVR has also achieved good results in many application fields, such as predicting natural disasters [4], [11], [28]. In Ref. [24], the author uses a particle swarm optimization algorithm (PSO) combined with a support vector machine (SVM) to predict the charging state of lithium battery in fault state through cross-test. The authors of [21] found the ideal coefficients for the four kernel SVR model’s kernel functions using the Human Learning Optimization Algorithm (HLO), and then utilized the ideal SVR model to simulate the compressive strength of concrete.
The newly suggested BOAPPE algorithm is the primary focus of this research. The SVR model’s unknown parameters are retrieved using real data set by this algorithm, and the SVR model is continuously refined to precisely measure the power output. Additionally, in order to train the SVR model to estimate the maximum power of the base-load power plant equipment, comparative experiments were conducted using BOA, PPE, PSO, Moth Flame Optimization Algorithm (MFO), and Aquila Optimizer (AO). The following are the contributions of this paper:
1. To optimize the SVR model, this study uses a better hybrid heuristic method called BOAPPE [8], [29], which combines BOA and PPE. It also uses two search techniques to increase the search space and speed up convergence while maintaining a balance between development and exploration.
2. In this study, six hybrid mathematical models are constructed, studied, and compared for their ability to predict power generation: BOA-SVR, PPE-SVR, PSO-SVR, MFO-SVR, AO-SVR and BOAPPE-SVR. According to the experimental findings, the BOAPPE-SVR model is more accurate than other SVR models.
3. This paper extracts the unknown parameters of the four SVR kernel functions as precisely as possible and shows that the Gaussian kernel function is the most effective kernel function for the SVR model.
4. Use the proposed BOAPPE-SVR model to successfully analyze the CCPP data set in the UCI machine learning library.
The remainder of the paper’s structure is the following: Section 2 introduces similar works of earlier studies. Section 3 describes the methodology and materials employed in this paper. Section 4 introduces the revised algorithm; Section 5 contains the experimental data and performance analysis. The conclusion and future work are presented in Section 6.
Section snippets
Related work
This section introduces the related work done by forerunners in CCPP and then suggests our creative model.
For unknown reasons, the forecast of power demand has always been a nonlinear problem, which brings some difficulties to the forecast of power demand. Because intelligent optimization algorithm has good optimization ability and can help nonlinear power demand forecasts become more accurate, more and more researchers train forecasting models through intelligent optimization algorithm.
Materials and methods
In order to make the introduction of better algorithms in the following section easier, this section covers the research concepts and techniques of this work, focusing primarily on the mathematical formulae and models utilized in the SVR, BOA, and PPE algorithms.
BOAPPE algorithm
The disadvantages of the BOA algorithm, such as its slow convergence speed and propensity to easily fall into local optimums, are improved in this part with the BOAPPE algorithm. The BOAPPE algorithm uses the grouping concept. All particles are evenly split into two pieces by the algorithm. The algorithm is optimized in the first section using the BOA algorithm. The random walk technique is employed in the first section to broaden the search space of viable solutions and prevent settling into a
Analysis of the results and discussion
In this section, the BOAPPE algorithm and the SVR model are combined, the specific processes of the BOAPPE-SVR model in the experimental process are introduced, and ultimately the BOAPPE-SVR model is compared with the comparison experiment, and the experimental findings are extensively analyzed and discussed.
Conclusion and future work
An enhanced BOA method, known as the BOAPPE algorithm, is proposed in this study together with the PPE algorithm. Due to the limitations of BOA algorithm creation and exploration, the BOAPPE algorithm uses two ways to increase the exploration space, hasten convergence, and integrate the BOAPPE algorithm with SVR. They are used to predict a base-load power plant’s production at full capacity when combined. A total of 9568 data points from the Turkish CCPP from 2006, of which 75% were used for
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Funding
None. No fundings to declare.
References (44)
- et al.
Spatial prediction of landslide susceptibility in western Serbia using hybrid support vector regression (SVR) with GWO, BAT and COA algorithms
Geosci. Front.
(2021) - et al.
A sequential model-based approach for gas turbine performance diagnostics
Energy
(2021) - et al.
Parallel fish migration optimization with compact technology based on memory principle for wireless sensor networks
Knowl.-Based Syst.
(2022) - et al.
Novel hybrid intelligence models for flood-susceptibility prediction: Meta optimization of the GMDH and SVR models with the genetic algorithm and harmony search
J. Hydrol.
(2020) - et al.
Development and multi-utility of an ANN model for an industrial gas turbine
Appl. Energy
(2009) - et al.
Performance prediction and optimization of an organic Rankine cycle (ORC) for waste heat recovery using back propagation neural network
Energy Convers. Manage.
(2020) - et al.
Two stage forecast engine with feature selection technique and improved meta-heuristic algorithm for electricity load forecasting
Energy
(2018) - et al.
A novel hybrid load forecasting framework with intelligent feature engineering and optimization algorithm in smart grid
Appl. Energy
(2021) - et al.
Short-term electricity price and load forecasting in isolated power grids based on composite neural network and gravitational search optimization algorithm
Appl. Energy
(2020) - et al.
Experiment on radial inflow turbines and performance prediction using deep neural network for the organic Rankine cycle
Appl. Therm. Eng.
(2019)