Short-term scheduling of gas-fired CHP plant with thermal storage using optimization algorithm and forecasting models

Highlights

• The concept of a computer-based tool for production optimization in the CHP plant of day-ahead planning is presented.

• The mathematical model of the gas-fired CHP plant was developed.

• The MLP (Multilayer Perceptron) forecasting model for electricity price was used.

• The artificial neural network was developed to predict the heat load in the DHN.

• The evolutionary method was used for solving the optimization problem.

Abstract

Accurate production planning is an important aspect of the combined heat and power plants operating on the electricity market. The complexity of production planning and scheduling depends mainly on the scale of the power system. Nowadays, the role of modern computer systems seems to be crucial and significantly affects optimal production planning in CHP plants. The production scheduling process must take into account the relationship between the production of heat and electricity in cogeneration units. An important aspect of optimal scheduling of power systems is precisely forecasting heat demand in district heating networks and electricity prices on the market.

In this paper, an optimization-based model for short-term scheduling of gas-fired CHP plant with heat accumulator is presented. The optimization model consists of a detailed simulation model of a cogeneration plant which is combined with an evolutionary algorithm. The optimization objective is to maximize the total gross margin for the day-ahead horizon of the CHP operation. An artificial neural network model is used for predicting heat demand in the district heating network. Different forecast models were tested for the electricity price forecast – extreme learning machines, multi-layer perceptron, auto-ARIMA, and triple exponential smoothing methods. The presented results show that the developed computer-based tool is efficient and effective for short-term scheduling of CHP plant with gas turbines and heat accumulator.

Introduction

The forecast process is substantial in many economic sectors, which are affected by different factors. The effectiveness of the management or planning mechanism directly results from the quality of generated forecasts. In the heat and power area, the accurate and reliable forecast of heat demand and electricity price has a significant impact on optimal production planning and effective operation of power systems. The relevant factors with an important impact on the forecast quality include the following aspects: weather conditions (temperature, wind, solar radiation), energy demand (season, day of the week, time of the day, the behaviour of consumers), fuel market (coal, gas and biomass price), economic and political situation in the world. The forecasting process can be divided due to the time horizon as follows [1]: short-term (day-week), medium-term (week-month), and long-term (month-years). The evolution and progress in heat demand and electricity price forecasting derive from different areas of science, e.g. mathematics, economics, engineering, statistics. Each branch of science has developed the forecasting methods for its needs demonstrating different approaches. Most of the techniques concentrate on short-term forecasting, mainly in one-day ahead models with an hourly time step. In literature, the medium- and long-term forecasting models are less frequently seen, especially for optimal operation planning of cogeneration systems [2], [3].

Forecasting models are used in many areas of human activity where the estimation of future conditions is important and useful, e.g. economics, weather forecasting, energy and power sector, stock market, transportation. The selection of an appropriate forecasting method depends mainly on the type of the studied phenomenon, available data, and factors affecting the prediction. Forecasting methods can be divided into three main groups: qualitative, quantitative, and combined methods. Qualitative forecasting methods are used in situations where the availability of numerical data is limited and forecasts are developed based on the knowledge and experience of specialists. The qualitative forecasting methods include consumer surveys, executive opinions, Delphi techniques, and salesforce polling. In the case where numerical data and knowledge of the forecasted phenomenon are available, quantitative methods can be applied. The quantitative forecasting methods are as follows: casual forecasting (regression analysis, econometric models, input-output models) and time series models (trend/pattern analysis, regression analysis, exploratory analysis). The combination of qualitative and quantitative forecasting methods is applied by using artificial neural networks and other deep learning models [4], [5].

Several different models, concepts, and approaches have been tested in the area of electricity price forecasting. In the literature, numerous attempts have been made to classify models used for the prediction of electricity prices. The general classification of the electricity price forecasting models proposed by Weron [6] identifies the following groups:

• fundamental methods (include basic information of physical and economic relationships which describes the correlation between production and trading of electricity, e.g. parameter-rich fundamental models, parsimonious structural models),

• multi-agent models (use a simulation model of the operation of electricity market including dependencies between market participants, the electricity price is determined by matching the supply and demand curves, e.g. Nash-Cournot framework, strategic production-cost models, agent-based simulation models),

• reduced-form models (provide information about the statistical properties of electricity prices over time, without indicating the exact value of hourly price forecasts, e.g. jump-diffusion models, Markov regime-switching models),

• statistical approaches (use mathematical models to determine the prediction of electricity prices based on the relationship between the historical data and other factors affecting the forecast, e.g. exponential smoothing methods, regression models, time series models),

• artificial intelligence techniques (utilize nature-inspired mathematical models to solve different tasks which are a problem for traditional algorithms, e.g. artificial neural networks, fuzzy models, support vector machines, and evolutionary algorithms).

The different methods and techniques are used for electricity price forecast, such as classical time-series models, artificial neural networks (ANN), fuzzy logic (FL), and support vector machines (SVM). It should be noted that most of the research studies present a combination of different techniques (from two or more) for electricity price forecasting. Amjady [7] proposed an efficient method based on a new fuzzy neural network for short-term price forecasting of electricity markets. The proposed method predicts hourly market-clearing prices for the day-ahead electricity markets. Zhang et al. [8] developed a new integrated model based on the improved empirical mode decomposition, autoregressive moving average with exogenous terms, exponential generalized autoregressive conditional heteroscedasticity, and adaptive-network-based fuzzy inference system. The obtained results show that the forecasting accuracy of the new integrated model proves higher than that of some well-recognized models in the literature. Osório et al. [9] presented a new hybrid evolutionary-adaptive methodology for electricity prices forecasting in the short-term time horizon. The proposed model is a combination of mutual information, wavelet transform, evolutionary particle swarm optimization, and the adaptive neuro-fuzzy inference system. Luo and Weng [10] proposed a two-stage supervised learning approach for electricity price forecasting. The proposed method is based on two types of mapping rules – first for mapping between the historical wind power and the historical price and the second for forecasting rule for wind generation. De Marcos et al. [11] presented a novel fundamental-econometric electricity price forecasting model. The proposed method combines a cost-production optimization (fundamental) model with an artificial neural network (econometric) model. In some researches, time series forecasting models are used. Contreras et al. [12] presented a method to predict next-day electricity prices based on the Auto-regressive integrated moving average (ARIMA) methodology. A detailed explanation of the ARIMA models and results from mainland Spain and Californian markets have been provided. Che and Wang [13] proposed a hybrid model that combines both support vector regression and ARIMA models to take advantage of the unique strength of SVR and ARIMA models in nonlinear and linear modelling. The experimental results demonstrate that the developed model outperforms the existing neural-network approaches, the traditional ARIMA models, and other hybrid models based on the root mean square error and mean absolute percentage error. Yang et al. [14] presented a new hybrid model for day-ahead electricity price forecasting, that combines the wavelet transform, the kernel extreme learning machine (KELM) based on self-adapting particle swarm optimization, and an autoregressive moving average (ARMA). The obtained results show that the proposed method is more accurate than individual methods and other hybrid methods.

In general, the forecasting models of heat demand in DHN use similar techniques and methods as for the electricity price prediction. Suryanarayana et al. [15] presented a novel method to forecast heat loads in district heating networks using automated feature selection with polynomial linear regression and deep learning techniques. Fang and Lahdelma [16] proposed a combination of SARIMA with linear regression to forecast the heat demand in a district heating network. The high accuracy of the developed model has been achieved for both short-term and long-term forecasts. Nielsen and Madsen [17] proposed a grey-box approach for the prediction of heat consumption in district heating systems. Xue et al. [18] developed a machine learning-based model for multi-step ahead DHN heat load forecasting. The support vector regression, deep neural network, and extreme gradient boosting (XGBoost) are used as the base learner to develop forecasting models. It can be stated that different methods and techniques have been tested for the prediction of electricity price and heat demand in district heating. The continuous development of computer technology stimulates the possibility of further research related to the possibility of increasing precision and accuracy of forecasting in the different fields of application.

Many different approaches to the problem of optimization of production planning in a CHP plant can be found in the literature. Tous et al. [19] developed a stochastic model to support decisions on production planning in a waste-to-energy plant on a short-term horizon. The obtained results showed that stochastic simulation-based planning provides increased revenues related to proper production planning. Daraei et al. [20] investigated the influence of poly-generation on the production planning of the CHP plant. The optimization problem is solved by using a Mixed Integer Linear Programming (MILP) model to minimize production cost in the studied system. The results confirm that proper production planning reduces fuel consumption and thus increases economic benefits. Wang et al. [21] developed an optimization-based method for planning and operating a CHP-based DH system. The optimization function is designed to minimize the overall costs of the net acquisition for heat and power in the deregulated power market. Results indicated that the developed method is efficient and flexible for planning and operating CHP-DH systems. Mitra et al. [22] presented a generalized component-based model for operational optimization of industrial CHP plants undertime-sensitive electricity prices. The results showed that the economic benefit of 5–20% depending on the level of utilization can be achieved. Zidan et al. [23] developed an optimization model for the production planning of CHP plants within microgrids. The multi-objective genetic algorithm (GA) was used to solve the planning problem. The optimization task was related to the minimization of the total net present cost and carbon dioxide emission.

The participation of CHP systems in the energy Day-ahead market requires accurate and efficient computer-based systems allowing for reliable planning of electricity and heat production. The optimal operation of CHP plants is a complex optimization problem that is influenced by many external factors such as weather conditions, heat demand in DHN, electricity price on the energy market, environmental and technical constraints. The proper planning of CHP operation needs powerful solutions that can propose the most beneficial production schedule for a specific period. Kumbartzky et al. [24] presented a comprehensive concept of an algorithm for increasing the profitability of CHP plant operation with heat storage by participating in multiple electricity markets. A multistage stochastic mixed-integer linear programming (MILP) methods are used in the optimization of the CHP system cooperation with heat storage and bidding in sequential electricity markets. Jin et al. [25] proposed a hybrid optimization method for the optimal day-ahead scheduling for the integrated urban energy system (IUES) including the CHP units. The goal function is to minimize the day-ahead operation cost of the IUES. The presented model is based on a genetic algorithm (GA) and a nonlinear interior point method (IPM) is utilized in optimization. Fang et al. [26] developed a stochastic optimal dispatch model of considering thermal and electrical coordination of the CHP plant. A detailed model of a district heating network is presented and the possibility of using the DHN as energy storage was quantitatively analysed. Kia et al. [27] proposed a novel method for optimal day-ahead scheduling of CHP units with electric and thermal storage systems. The mixed-integer non-linear (MINLP) model coupled with linearization techniques was used in the optimal scheduling of CHP units. Besides the literature review, there are not many commercial software tools that allow scheduling the production in CHP plants [28], [29], [30]. The combination of a simulation model of energy systems with forecasting models of electricity price and heat demand in DHN for optimization of the day-ahead CHP operation is a novel approach and rarely observed the solution.

The planning process of a combined heat and power (CHP) plant operation on the electricity market is characterized by high uncertainty related to predicting future electricity prices and heat demand in the district heating network over a specific time horizon. This means that every decision concerning the scheduling of the production unit is made in an unknown market environment. For CHP plants equipped with gas turbines, the process of planning electricity and heat production strictly dependent on weather conditions. An important aspect of the process of planning the operation of cogeneration systems is to have the appropriate engineering tools to precisely determine the production of different energy carriers. Combined heat and power plants, as participants in the electricity market, are required to report their projected production of electricity supplied to the grid. This situation forces the CHP plant to prepare a plan for production units to maximize future profits or minimize production costs. The scheduled production profile of the CHP plant must take into account the heat demand of the district heating network, electricity price forecast, environmental aspects, and technical constraints. The proper forecasting of heat demand in the district heating network is crucial for the accurate planning of the CHP plant operation on a short-term horizon. The availability of a heat storage tank allows flexible adaptation of the production units to maximize electricity production at times when the electricity price is high. To develop the production plan for a day-ahead period, it is necessary to make forecasts concerning predicted electricity prices and heat demand in the district heating network. Based on these two forecasts, it is possible to select an optimal configuration of production units.

In this paper, an efficient and practical approach for the planning of CHP operation on the Day-ahead market is presented. The concept of a computer-based model for the optimization of the production planning process in the CHP plant is developed. Optimal production planning and operation of the CHP plant on a power market requires considering both forecasts of electricity prices and the heat demand. Both heat demand and power prices fluctuate hourly and seasonally. Therefore, the planning tool has been equipped with modules for forecasting these parameters. Based on the literature review, it can be concluded that most of the studies do not solve the problem of optimal production planning comprehensively. The proposed tool is a combination of a forecasting model of electricity prices, a predictive algorithm of heat demand in DHN, a mathematical model of CHP plant, and an optimization algorithm for scheduling the operation of heat storage and production units. The objective is to maximize the total gross margin for the day-ahead horizon of the CHP operation. The optimization problem is solved using the evolutionary algorithm. The proposed methodology combines available methods of mathematical modelling, forecasting, and optimization techniques into a single solution that supports the planning process for CHP plants. The paper is organized as follows: Section 1 gives a summary of a literature review about different techniques and methods used in forecasting electricity price and heat demand in DHN. Section 2 provides the basic information about the Polish Day-ahead market as well as the description of the electricity price forecasting methods used in the calculation. The forecast results for different models are presented. Sections 3 presents a concept of a computer-based tool for optimization of the production planning process, a simulation model of CHP plant, and a predictive model of heat demand in DHN with the simulation results, respectively. Section 4 shows the calculation example of the optimization of the production plan for the selected day-ahead horizon. The last section provides a conclusion and a summary of the obtained results.