Sage Journals: Discover world-class research

Abstract

In view of the uncertainty and volatility of wind power generation and the inability to provide stable and continuous power, this paper proposes a hydrogen storage wind-gas complementary power generation system, using Matlab/Simulink to simulate and model wind generators and gas turbines. Considering the economy and power supply reliability of the wind-gas complementary power generation system, and taking the economic and environmental cost of the system as the objective function, the capacity optimization model of the wind-gas complementary power generation system is established. The brain storming algorithm (BSO) is used to solve the optimization problem, and the BSO algorithm is used to optimize the BP neural network, which improves the accuracy of the BP neural network for load forecasting. Finally, a simulation is carried out with load data in a certain area, and the simulation verification verifies that BSO-BP can improve the accuracy of load forecasting and reduce the error of load forecasting. Multi-objective optimization of system economic cost and environmental cost through BSO algorithm can make the system cost reach the most reasonable level. Through the analysis of the calculation examples, it is verified that gas-fired power generation can effectively alleviate the volatility of wind power generation, showing the role and advantages of energy complementary power generation. Therefore, the wind-gas complementary system can effectively increase energy utilization and reduce wind curtailment.

Keywords

Wind power generation gas power generation load forecasting brain storming algorithm BP neural network

Introduction

In order to cope with environmental pollution and energy shortages, China is now developing rapid new energy technologies. Among them, wind power has developed particularly rapidly. In 2017, China's total installed wind power capacity was 164 million kW, an increase of 10.5%, accounting for 9.2% of total wind turbine capacity (National Energy Administration of the People's Republic of China, 2018). It is undeniable that wind power also has some shortcomings. Wind power itself has volatility and untainty (Li, 2012), which makes it difficult to be completely absorbed. In 2017, 41.9 billion kW⋅h of electricity generated by wind power in China was not used, and the average abandonment rate was as high as 12% (National Energy Administration of the People's Republic of China, 2018). The wind power consumption situation is more severe, and its large-scale grid connection will cause the grid to fluctuate and it will not be able to provide electricity normally, which has huge hidden danger. When the wind power permeability increases to a certain extent, only thermal power is used to deal with the uncertainty of wind power output. Frequent start and stop of thermal power units will cause fluctuations in the power grid and cannot operate safely and stably.

Gas turbine, as a high operational flexibility equipment in power system, its output power can quickly track the fluctuation of wind power output, which is expected to alleviate the phenomenon of wind abandonment in power system. As of 2017, China’s gas turbine assembled machines reached 76.29 million kW, accounting for 4.29% of the installed power generation capacity, and the annual power generation capacity reached 220 billion kW⋅h. In the coastal areas of Jiangsu, Zhejiang, Shanghai and Guangdong, gas engine assembly accounts for more than 30% (Liu and Wang, 2018). The China Energy Department issued the “Thirteenth Five-Year Plan for Electric Power Development”, proposing to increase the ratio of natural gas utilization and orderly develop natural gas power generation, by then, the application of electricity and natural gas will be closer (National Development and Reform Commission of the People's Republic of China, 2016). Complementary power generation is a way to increase energy utilization in the context of current energy shortages, and is widely used in the field of distributed new energy (Xiong et al., 2015). At present, the world is based on the complementary characteristics of different primary energy. In the production process of secondary energy, through reasonable coordination and cooperation, it can achieve good energy efficient use and stable supply of electrical energy. Wind-solar complementary system is one of the most in-depth complementary systems in research. Literature (Benlahbib et al., 2020) proposed a hybrid microgrid system based on wind and solar power generation for remote area applications. Through the control of power electronic devices to improve the utilization of power, but there is no mutual influence between the systems. Literature (Tan et al., 2021) proposes a wind-solar-water hybrid power generation system, which uses different energy sources to complement each other, reduces the impact of wind and solar fluctuations on electric energy, and improves the quality of power output from the grid. Since the influencing factors in the multi-energy complementary system are more complicated, a unilateral increase in the efficiency of energy production cannot improve the overall coordinated operation of the system. Therefore, literature (Bouchekara et al., 2021) uses a multi-objective method to optimize the design of the microgrid, and obtains the optimal solution according to the needs to maximize the system economy.

The above-mentioned complementary models are mainly based on wind-solar complementation, and there are few domestic and foreign researches on the combination of wind and natural gas. The national natural gas ownership is low and unevenly distributed, the natural gas reserves are insufficient, and the network infrastructure is relatively backward. In 2017, China needed a total of 230 billion m³ of natural gas for production and living, but the gas production was only 138 billion m³, and the natural gas supply was insufficient. In addition, according to China’s specific conditions, China’s residential gas has a higher priority than non-residential gas industries (Sun et al., 2018). When the natural gas supply is insufficient, priority will be given to limiting the natural gas supply of gas-fired units, thus limiting the variation range of their output power, thereby reducing the wind power consumption capacity of the power system. Therefore, this is also a factor restricting the development of gas power generation technology in China.

Due to gas turbine has the characteristics of high operation flexibility and rapid response, Therefore, gas turbines as a supplementary energy source have broad prospects for solving the adverse effects of wind and photovoltaic power generation. Literature (Chen et al., 2020) established a hotspot co-generation model to combine natural gas with wind energy and solar energy to increase the power generation and energy efficiency of renewable energy. In reference (Yao, 2020) Established a short-term coordination plan for electricity and natural gas and a real-time dispatch system to predict short-term wind power changes to reduce forecast errors and reduce the impact of energy uncertainty on air pressure changes. But when the output of wind power is insufficient, it cannot provide enough power to the grid.

Therefore, this article aims at the problem that wind power cannot provide stable power, and builds a wind-gas complementary power generation system, the wind turbine, gas turbine and electrolyzer are modeled by Simulink and multi-objective optimization. The BSO algorithm is used to solve the multi-objective optimization model, and the economic and environmental factors of the wind and air complementary system are analyzed, which can significantly reduce the investment cost of the system. BSO algorithm is used to improve BP network, which improves the prediction accuracy of BP network, and compare the load forecast results with the output of wind power and gas power generation. The wind-gas complementary power generation system is proved to be able to effectively improve the volatility of wind power generation, improve the power quality, and the energy can be fully utilized. The analysis results further prove the rationality of the model and the superiority of BSO-BP network algorithm.

Model introduction

Wind-gas complementary power generation system structure

The complementary power generation system composed of wind generators, micro gas turbines, AC/DC converter, electrolyzers and other equipment connected to the grid can provide electrical energy for the loads in the entire region. The system is shown in Figure 1. The excess power enters the electrolyzer to electrolyze water and store hydrogen.

Figure 1.

System structure diagram.

Dynamic mathematical model of micro gas turbine

Compressor

The micro gas turbine studied in this paper uses a centrifugal compressor. The characteristics of the compressor are usually described by the compressor performance curve. The compressor characteristic curve is expressed by the equivalent flow $G_{c}$ , the equivalent speed $N_{c}$ , the compressor pressure ratio $π_{c}$ , the compressor efficiency $η_{c}$ and other parameters (Feng, 2019) $\bar{G} = \frac{G \sqrt{T}}{P}$ (1) $\bar{n} = \frac{n}{\sqrt{T}}$ (2)

Pressure ratio and efficiency are the combined flow rate $\bar{G}$ and rotation speed $\bar{n}$ , which can be approximated by fitting the compressor characteristic curve $π_{c} = f_{1} (\bar{G}, \bar{n})$ (3) $η_{c} = f_{2} (\bar{G}, \bar{n})$ (4)

Volumetric inertia

A volumetric inertia model was developed considering the instability of flow caused by the pipeline between the compressor and the combustion chamber and the internal combustion chamber (Zhong and Yan, 2019). Using the lumped-parameter method, assume that the pressure loss is concentrated at the outlet $\frac{d p_{out}}{d t} = \frac{R T_{out}}{M_{a} V} (g_{a, i n} - g_{a, out})$ (5)where $T_{out}$ and $p_{out}$ are outlet temperature and pressure respectively; $g_{a, i n}$ and $g_{a, out}$ are inlet and outlet air flow respectively; $M_{a}$ is air molar mass.

Combustion chamber

The combustion chamber is a component with strong heat transfer characteristics. The imbalance of energy input and output will cause temperature change, so the thermal inertia of flue gas in combustion chamber is considered. The dynamic equation describing the flue gas temperature at the combustor outlet is obtained from the unsteady energy balance equation $\frac{d (m_{c c} u_{c c})}{d t} = g_{a, i n} h_{a, i n} + g_{f} (h_{f} + η_{c c} L H V) - g_{g, out} h_{g, out}$ (6)where $m_{c c}$ and $u_{c c}$ are the internal energy of flue gas mass and unit mass of flue gas in the combustion chamber; $h_{a, i n}$ , $h_{f}$ and $h_{g, out}$ are the enthalpies of air, fuel and flue gas at the inlet of the combustion chamber; $LHV$ is the net specific energy of fuel; $η_{cc}$ is the combustion efficiency (Yan et al., 2014). The dynamic equation of flue gas outlet temperature change can be obtained $τ_{c c} \frac{d T_{out}}{d t} = \frac{g_{a, i n} h_{a, i n} + g_{f} (h_{f} + η_{c c} L H V) - g_{g, out} h_{g, out}}{g_{g, out} c_{p, g}}$ (7)where $T_{out}$ is the outlet temperature of flue gas in combustion chamber; $τ_{c c}$ is the time constant, which can be expressed as $τ_{c c} = \frac{m_{c c}}{k_{g} g_{g, out}}$ (8)where $k_{g}$ is the adiabatic index of flue gas.

Turbine

Centripetal turbine is widely used in micro gas turbine. Like centrifugal compressor, the reduced flow rate $\frac{g_{g} \sqrt{θ}}{δ}$ and isentropic efficiency $η_{i s, T}$ of turbine are obtained by linear interpolation of characteristic curve. The turbine exhaust temperature $T_{out}$ and output power $P_{T}$ are calculated by the following formula $T_{out} = T_{i n} [1 - (1 - β_{T}^{- \frac{k_{g} - 1}{k_{g}}}) η_{i s, T}]$ (9) $P_{T} = g_{g} c_{p, g} T_{i n} (1 - β_{T}^{- \frac{k_{g} - 1}{k_{g}}}) η_{i s, T}$ (10)

Rotor

Without considering the load characteristics of the turbine, only the output power is considered, the inertia equation of rotor rotation is as follows $\frac{d n}{d t} = \frac{1}{4 π^{2} J} \frac{1}{n} (P_{T} - P_{C} - P_{L})$ (11)where $J$ is rotor moment of inertia; $P_{T}$ , $P_{C}$ and $P_{L}$ are turbine output power, compressor power consumption and load power respectively.

Wind power generation model

The aerodynamic model of wind turbine is (Kong et al., 2021) $P_{m} = \frac{ρ_{air} C_{p} (λ, β) π R^{2} υ_{w}^{3}}{2}$ (12)where $P_{m}$ is the mechanical power of the wind turbine converted from the energy captured by the wind turbine, $ρ_{air}$ is the air density, $R$ is the radius of the wind turbine impeller, $λ$ is the tip speed ratio, $β$ is the pitch angle, $C_{p}$ is the wind energy conversion efficiency coefficient of the blade, and $υ_{w}$ is the wind speed.

Alkaline electrolyzer model

The $U - I$ equation of the cell is (Zhang et al., 2020) $U_{cell} = \frac{Δ G}{z F} + \frac{r_{1} + r_{2} T_{e l}}{A_{cell}} I_{e l} + (s_{1} + s_{2} T_{e l} + s_{3} T_{e l}^{2}) \log (\frac{t_{1} + t_{2} / T_{e l} + t_{3} / T_{e l}^{2}}{A_{cell}} I_{e l} + 1)$ (13)where $Δ G$ is the Gibbs free energy change of the electrochemical reaction process, $z$ is the number of electron transfer per reaction, $F$ is the Faraday constant, $r_{1}$ and $r_{2}$ are the ohmic resistance parameters of the electrolyte, $T_{e l}$ is the temperature of the electrolytic cell, $A_{cell}$ is the area of the electrolytic module, $I_{e l}$ is the DC current, $s_{n}$ is the electrode overvoltage coefficient (n = 1…3) and $t_{n}$ is the electrode overvoltage coefficient (n = 1…3).

The hydrogen production rate of alkaline electrolyzer is ${\dot{n}}_{H_{2}} = η_{F} \frac{N_{e l} I_{e l}}{z F}$ (14)where the expression of $η_{F}$ is $η_{F} = a_{1} \exp (\frac{a_{2} + a_{3} T_{e l}}{\frac{I_{e l}}{A_{cell}}} + \frac{a_{4} + a_{5} T_{e l}}{\frac{I_{e l}}{A_{cell}}})$ (15)where, $N_{e l}$ is the number of modules in series, $z$ is the number of electron transfer in each reaction, and $a_{n}$ is the Faraday efficiency coefficient (n = 1…5)

The overall power consumption of the electrolytic cell can be expressed as $P_{e l} = N_{e l} U_{e l} I_{e l}$ (16)where $U_{e l}$ is the end voltage of the electrolytic cell.

Introduction of BSO algorithm

Brief introduction of algorithm

The process of brainstorming is a creative way for people to solve problems. The process of solving problems is to gather people from different professional fields for divergent thinking, collision of ideas and fusion of ideas, and finally find the most suitable solution. BSO algorithm is an optimization algorithm obtained by abstracting the brainstorming process. BSO algorithm is obtained by abstracting the various elements and processes in the brainstorming process. Each idea in the brainstorming process is abstracted in the algorithm as an individual in the optimization problem solution space, and the set of individuals is represented as a viable solution to the optimization problem. The algorithm abstracts the different stages of the brainstorming process into four operations: clustering, mutation, generation of new individuals, and selection. Four operations are used to update individuals, and ultimately the best solution to the problem is found (Chen et al., 2019b).

Brain storming algorithm is mainly composed of clustering and mutation. K-means clustering algorithm is a BSO clustering method, similar individuals are grouped together as a group, forming a total of k categories, and takes the optimal individual of fitness function value set by itself as the clustering center. At the same time, there is a certain probability that a new population individual will replace one of the cluster centers to avoid falling into the local optimum.

There are four main variation modes of BSO:

Randomly select the best individual in a class and add a disturbance variable to form a new population mutant individual

Arbitrarily select individual elements in a population to add random interference to form a new population element;

Fusion of two random class centers and randomly adding disturbances to generate a new population individual;

Pick two classes arbitrarily from all classes, merge any two individuals in the two classes and add random disturbance to generate a new mutant individual.

The probability that the most prominent individual in each cluster center is selected is $p_{j} = \frac{| M_{j} |}{N}$ (17)where $| M_{j} |$ is the number of individuals in class j.

The new individual generation formula is as follows $x_{n d} = x_{s d} + ξ * N {(0, 1)}_{d}$ (18) $ξ = \lg sig ((0.5 * T - t) / k) * R (0, 1)$ (19)where $x_{n d}$ is the new d-dimensional individual, $x_{s d}$ is the selected individual, $T$ and $t$ represent the maximum number of iterations set and the current number of iterations respectively, $k$ can adjust the slope of $\lg sig (\cdot)$ function, $N {(0, 1)}_{d}$ is the d-dimensional standard normal distribution, and $R (0, 1)$ is a random value of 0 ∼ 1.

BSO algorithm flow

Initialization: n individual X_i as the initial solution of the optimization problem are randomly selected from all feasible solutions, where i = 1, 2 , … , n. As the maximum number of iterations of the algorithm, I_max sets the algorithm termination condition.

Clustering: Divide n individuals into m classes and use the K-means aggregation method to calculate the function value corresponding to the fitness of n individuals. The individual with the best function value is recorded as the center individual of the class.

Mutation: Randomly generate a random number p₀₁ between [0, 1]. If it is greater than the preset probability parameter p₀, not only re-select the newly generated individual for replacement of a class but also let the randomly generated new individual replace it.

Generate new individuals: first generate the random number p₁₁ between [0, 1], compare with the pre-set probability parameter p₁, and select the following methods to generate candidate individuals X_S according to the results.

If p₁₁ > p₁, a class is randomly selected to generate the random number p_a1 between [0, 1] and compare it with the preset probability parameter p_a:

if p_a1 > p_a, then the selected category as a candidate center individual subject X_s;

Otherwise, randomly select class non-central individuals as candidate individuals X_s.

If p₁₁ ≤ p₁, select 2 classes at random, generate a random number p_b1 between [0, 1], and compare it with the preset probability parameter p_b:

If p_b1 > p_b, select 2 class-centered individuals and mark them as individuals X_s1 and X_s2;

Otherwise, two non central individuals are randomly selected from the two classes, which are denoted as X_s1 and X_s2.

Fuse individual X_s1 and X_s2 according to the formula to get X_s $X_{s} = ω \times X_{s 1} + (1 - ω) \times X_{s 2}$ (20)where ω is a real number in the interval [0, 1], The candidate X_s individual generates a new population individual Y after adding random interference $Y = X_{S} + ξ \times N (μ, σ)$ (21)where N (μ, σ) is the Gaussian random function with the center at μ and the variance at σ; $ξ$ is the coefficient of Gaussian random function to adjust the amplitude of random disturbance. $ξ$ is usually a logarithmic S-transform function in the form of $ξ = \log sig ((0.5 \times I_{\max} - I_{c}) / k) \times rand (0, 1)$ (22)where I_max is the number of iterations of the algorithm; I_c is the number of iterations of the current algorithm; k controls the slope of the log sig function; the function of rand (0, 1) is to generate a random between (0, 1) number.

5. Individual selection: The fitness of the new individual Y is compared with the currently updated individual. Individuals with good fitness enter the next generation. After all n individuals are updated, go to 6), otherwise go to 4) to update the individual;

6. If the individual reaches the optimal solution condition or reaches the preset number of algorithm iteration termination times, the algorithm stops running, otherwise go to 2) Start a new iteration process (Chen et al., 2019a).

Bso-BP neural network prediction model

In the fields of pattern recognition, signal processing, etc., the influence of BP neural network is quite large, but there is still a difficult problem in the design of the network, that is, the determination of the structure. The weight value and threshold value of the BP neural network are randomly assigned. This paper uses the BSO algorithm to optimize, after training, the optimized BP neural network is used to predict the load (Liang et al., 2019). The algorithm flow of BSO-BP neural network is shown in Figure 2.

Figure 2.

BSO-BP algorithm flow chart.

Figure 3.

The fitness convergence curve of the three algorithms.

System optimization configuration model

Objective function

The objective function is established with the minimum economic cost of system operation and minimum environmental pollution as the optimization objective (Zhao et al., 2021; Zhu and Liu, 2017) $\min F = \sum_{t = 1}^{T} (F_{1} + F_{2})$ (23)where F is the general contract for microgrid and grid-connected operation, F₁ is the economic cost, F₂ is the environmental cost, and T is time scheduling.

Objective function 1: Economic cost

For the microgrid, the main expenditures come from the operation and maintenance costs of the gas turbine and the cost of fuel consumption, as well as the cost of interaction with the electric power of the large grid $\min F_{1} = C_{1} (t) + C_{2} (t) + C_{3} (t)$ (24)where $C_{1}$ is the combustion cost. In this article, the combustion cost is the combustion cost $C_{M T}$ of the gas turbine. $C_{2}$ is the operation and maintenance cost; $C_{3}$ is the power grid interaction cost.

Operation and maintenance cost $C_{2}$ $C_{2} = λ_{1} P_{W T} (t) + λ_{2} P_{M T} (t)$ (25)where $λ_{1}$ and $λ_{2}$ are the wind turbines respectively, and the unit price of maintenance cost of the gas turbine: $P_{M T}$ and $P_{W T}$ are the output power of the the gas turbine and wind turbine in the $t$ period.

Power grid interaction cost $C_{3}$ $C_{3} = \sum_{t = 1}^{T} (C_{buy} (t) - C_{set} (t))$ (26) $C_{buy} (t) = r_{buy} (t) \cdot P_{buy} (t)$ (27) $C_{sel} (t) = r_{sel} (t) \cdot P_{sel} (t)$ (28)where $r_{buy} (t), r_{sel} (t)$ are the unit price of electricity purchased and sold by the system to the power network in $t$ period respectively $C_{buy} (t), C_{sel} (t)$ are respectively the cost of purchasing and selling electricity from the grid during the period t; $P_{buy} (t), P_{sel} (t)$ are the power that the system buys and sells from the grid during the $t$ period.

Objective function 2: Environmental cost

Minimal cost of system environmental pollution control $\min F_{2} = \sum_{t = 1}^{T} (\sum_{i = 1}^{I} (v_{i} (\sum_{j = 1}^{N} β_{j i} P_{j}^{t} + β_{gridi} P_{grid}^{t})))$ (29)where F₂ is the pollutant treatment cost of the system; $β_{gridi}$ is the emission coefficient of the $i$ pollutant from the grid. $P_{j}^{t}$ is the output power of the $j$ output source; $P_{grid}^{t}$ is the output power of the main grid; $v_{i}$ is the treatment cost of the $i$ pollutant, Including SO₂, CO₂, NOX, etc. $β_{j i}$ is the emission coefficient of the $i$ pollutant of the $j$ output source (Huang et al., 2020; Xu et al., 2018).

Because the optimal economic cost and the optimal environmental cost are contradictory, the linear weighting method is used to express the comprehensive benefits as $\min F_{3} = ω_{1} F_{1} + ω_{2} F_{2}$ (30)where $ω_{1}$ is the proportion of economic costs, and $ω_{2}$ is the proportion of environmental costs. F₁ and F₂ are two mutually restrictive variables. When the economic cost is reduced, the environmental cost will rise, and when the environmental cost is reduced, the economic cost will rise again. Therefore, the satisfaction model is established as ${\begin{cases} X = \frac{F_{1} - F_{1}^{\min}}{F_{1}^{\max} - F_{1}^{\min}} \\ Y = \frac{F_{2} - F_{2}^{\min}}{F_{2}^{\max} - F_{2}^{\min}} \end{cases}$ (31)where $X, Y \in [0, 1]$ , 0 means satisfaction and 1 means dissatisfaction.

Constraints

Power balance constraints (Gao et al., 2020; Li and Li, 2018) $P_{load}^{t} = \sum_{i = 1}^{N} P_{G i}^{t} + P_{grid}^{t}$ (32)

where

P_{load}^{t}

is the total power required by the load in the system during period

t

;

P_{G i}^{t}

is the total power emitted by various output sources during

t

period;

P_{grid}^{t}

is the interactive power between the system and the grid during

t

period.

2. System and grid interaction power constraints $P_{grid}^{\min} \leq P_{grid}^{t} \leq P_{grid}^{\max}$ (33)

where $P_{grid}^{\min}$ and $P_{grid}^{\max}$ represent the minimum and maximum of the interactive power between the system and the grid, $P_{grid}^{t} < 0$ means the system delivers power to the grid.

3. Constraints on the output of each output source $P_{i}^{\min} (t) \leq P_{i} (t) \leq P_{i}^{\max} (t)$ (34)

where

P_{i}

is the true output power of output source

i

during

t

period;

P_{i}^{\min} (t)

and

P_{i}^{\max} (t)

are the limits of the output power of the output source

i

during the t period.

Case analysis

Taking the load change in a certain area in the northern winter as the research object, selecting the load data of 24 hours a day for a period of time and predicting with a group of ten days, setting 1 h as the sampling time node for data collection and sample training. The parameter settings are shown in Table 1.

Table 1.

Parameters of the system.

Parameters	Value
Population size	80
BP neural network learning rate	0.1
Mutation probability	0.2
Training times	1000
The maximum number of iterations	80
BP neural network structure	1-7-1

The NSGA-II algorithm is a fast non-dominated sorting genetic algorithm. It is one of the commonly used algorithms for solving multi-objective optimization problems. It belongs to the group intelligence optimization algorithm like the brainstorming algorithm (BSO) and the particle swarm optimization algorithm (PSO). This paper verifies the advancement and rationality of the algorithm proposed in this paper by comparing the BSO algorithm with the NSGA-II algorithm and the PSO algorithm under the same conditions.

Figure 3 shows the convergence curves of the three algorithms. In the figure, the BSO algorithm begins to converge at the earliest and reaches the minimum when the number of iterations is 4. The fitness value of the NSGA-II algorithm reaches the minimum when the number of iterations is 10, and the fitness value of the PSO algorithm reaches the minimum when the number of iterations is 40. Therefore, it can be known that the BSO algorithm has the fastest convergence speed and can obtain the optimal solution to the problem in the shortest time. The convergence speed of the NSGA-II algorithm is lower than that of the BSO algorithm, and its convergence effect and solution ability are poor. The PSO algorithm has the worst convergence effect and the lowest solution ability.

After optimizing BP neural network with BSO algorithm, the load data collected in recent years are trained and predicted. The result is shown in Figure 4. The trend in the figure shows that the fitting degree between the predicted change trend and the actual change trend is very high. From the error distribution diagram in Figure 5, it can be found that the error of each prediction point does not exceed 0.08, which has high prediction accuracy. Therefore, the BSO-BP algorithm has higher forecasting accuracy and smaller forecasting errors for load forecasting.

Figure 4.

BSO-BP neural network load forecasting curve.

Figure 5.

Load forecast error curve.

The system configuration of each component of the system is shown in Table 2.

Table 2.

System configuration.

Name	Installed power	Quantity	Costs
Wind turbines	3.4 MW	2	23,700¥
Gas turbine	10 MW	1	13,060¥
Electrolyzer	400 KW	10	15,000¥

The working parameters of the electrolytic cell are shown in Table 3.

Table 3.

Working parameters.

Parameter	Value	Unit
Current density	0.1–0.2	A/cm²
Voltage range	2.04–2.14	V
Hydrogen production	≤80,000	Nm³/h
Power consumption	5.0	kWh/Nm³
Operating temperature	50–100	°C
Operating pressure	0.01–0.1	MPa
Effectiveness	75–90	%

It can be seen from Table 3 that there are many influencing factors in the hydrogen production process of the electrolytic cell. Therefore, in the modeling and simulation process, reasonable values are selected according to the range of each parameter in the table to simulate the working process of the electrolytic cell.

Take typical northern winter days as the analysis object, it is divided into two scenes for comparative analysis as shown in Table 4.

Table 4.

Scene setting.

Scene	Content
Scene 1	The gas turbine runs 24 h a day
Scene 2	The gas turbine runs in the morning, midnight and evening

It can be seen from Figure 6 that when the economic cost is the smallest, the gas turbine has a larger output. When the wind power output capacity is insufficient to meet the load demand, the gas turbine generates power to compensate for the required power. When considering environmental factors, as shown in Figure 7, the gas turbine cannot be operated for a long time. It only 7:00–8:00, 11:00–14:00 and 18:00–20:00 can run power generation to supplement the insufficient wind power output. Among them, the wind speed in the time period of 11:00–18:00 is relatively low, and the output power cannot meet the local load demand. It can be seen in Figure 6 that when the output capacity of wind power is insufficient, the electric energy generated by the gas turbine can make up for the insufficient output of wind power to meet the load demand. As shown in Figure 7, due to environmental cost factors, the gas turbine cannot be operated continuously for a long time. It only runs at a specific time in the morning, midnight and evening, and can play a role in compensating wind power during its operation time. In the time period of 0:00–6:00 and 15:00–17:00, the gas turbine stops working. At this time, the wind speed is low, and the wind power output cannot meet the load demand. Therefore, the energy management center needs to distribute electric energy from the grid to the load.

Figure 6.

Scenario 1 system output diagram.

Figure 7.

Scenario 2 system output diagram.

For the system to purchase electricity from the grid, as shown in Figure 8, the required value for scenario 1 is much lower than scenario 2. Although the scenario 2 reduces the working time of the gas turbine and relatively reduces the emission of pollutants, the reduction in environmental costs due to the increase in power purchase does not reduce the overall cost of the system, but increases the cost of the system. When the power generation of the system can meet the load demand, the system will send the excess power to the electrolytic cell device, electrolyze water in the electrolytic cell to produce hydrogen, and store the electrical energy in the form of chemical energy. By observing Figure 9, it can be found that the input power and working time of the electrolytic cell in scenario 1 are much greater than the input power and working time of the electrolytic cell in scenario 2. Therefore, more hydrogen energy can also be generated in scenario 1, which improves energy utilization.

Figure 8.

Comparison of power purchase curves in different scenarios.

Figure 9.

Electrolytic cell power curve in different scenarios.

Conclusion

This paper takes wind power and gas complementary systems as the research object, and proposes a load optimization algorithm based on BSO-BP. The energy management and load forecast of wind power and gas complementary systems are studied with the goal of minimizing the system cost. Conclusion as below:

The BSO algorithm is used to solve the multimodal high-dimensional function problem to optimize the initial weights and thresholds of the BP neural network, to improve the network performance, to make it converge quickly to a better solution, to improve the accuracy of load forecasting and to reduce the error of load forecasting.

Through multi-objective optimization calculation and analysis of economic costs and environmental costs, it is proved that the BSO algorithm can make the total cost of the system reach the most economical level, and make the benefit of the system reach the best.

After analysis of calculation examples, it is found that gas-fired power generation can effectively make up for the insufficiency of wind power generation at low wind speeds, and minimize the fluctuation of the power grid. The wind-gas complementary system can effectively increase energy utilization and reduce wind curtailment.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research,authorship,and/or publication of this article: This work is funded by the National Natural Science Foundation of China (51877070,51577048);funded by the Natural Science Foundation of Hebei Province (E2018208155);funded by the Talent Engineering Training Support Project of Hebei Province (A201905008);Project of Key R&D Plan of Hebei Province (20314501D,19214501D).

ORCID iD

Zheng Li

References

Benlahbib

Bouarroudj

Mekhilef

, et al. (2020) Experimental investigation of power management and control of a PV/wind/fuel cell/battery hybrid energy system microgrid. International Journal of Hydrogen Energy 45(53): 29110–29122.

Bouchekara

HRE

Javaid

Shaaban

, et al. (2021) Decomposition based multiobjective evolutionary algorithm for PV/wind/diesel hybrid microgrid system design considering load uncertainty. Energy Reports 7: 52–69.

Chen J, Deng C, Peng H, et al. (2019a) Enhanced Brain Storm Optimization with Role-playing Strategy. Proceedings of the 2019 IEEE Congress on Evolutionary Computation (CEC 2019). IEEE, 2019, pp. 1132–1139.

Chen

Fan

, et al. (2020) Energy flow optimization method for multi-energy system oriented to combined cooling, heating and power. Energy 211: 118536.

Chen XR, Li JQ, Han YY, et al. (2019b) An Improved Brain Storm Optimization for a Hybrid Renewable Energy System. IEEE Access 7: 49513–49526.

Feng

(2019) Research on modeling and predictive control of micro gas turbine. Southeast University, Bangladesh.

Gao

Wei

, et al. (2020) Research on wind power prediction based on optimal combination of NSGA-II algorithm. Hydropower 46(02): 114–118.

Huang

Cai

Xiao

, et al. (2020) Research on multi-objective optimal dispatching of grid-connected microgrid based on NSGA-II improved GSO algorithm. Journal of Sichuan University of Light Chemical Technology (Natural Science Edition) 33(06): 24–31.

Kong LG, Yu JM, Cai GW, et al. (2020) Power regulation of off-grid electro-hydrogen coupled system based on model predictive control. Proceedings of the CSEE, pp. 1–12. Available at: https://kns.cnki.net/kcms/detail/11.2107.TM.20200927.1155.004.html

10.

(2012) China wind power development report 2012. Beijing: China Environmental Science Press.

11.

LM and Li Z

(2018) Microgrid energy management based on hierarchical multi-objective optimization algorithm. Power Construction 39(04): 75–82.

12.

Liang

Guo

Zhu

(2019) BP neural network fuzzy image restoration based on brain storming optimization algorithm. Journal of Electronics and Information 41(12): 2980–2986.

13.

Liu

Wang

(2018) Development status and trend of gas power generation in China. International Petroleum Economy 26(12): 43–50.

14.

National Development and Reform Commission of the People's Republic of China (2016) The 13th five-year plan for renewable energy development. National Development and Reform Commission of the People's Republic of China.

15.

National Energy Administration of the People's Republic of China (2018) Wind power grid operation in 2017. National Energy Administration of the People's Republic of China.

16.

Sun

Zhang

, et al. (2018) Supply and demand situation, existing problems and policy suggestions of natural gas in China. Energy of China 40(3): 41–43, 47.

17.

Tan

Wen

Sun

, et al. (2021) Evaluation of the risk and benefit of the complementary operation of the large wind-photovoltaic-hydropower system considering forecast uncertainty. Applied Energy 285: 116442.

18.

Xiong

Wang

, et al. (2015) An optimization coordination model and solution for combined cooling, heating and electric power systems with complimentary generation of wind, PV, gas and energy storage. Proceedings of the CSEE 35(14): 3616–3625.

19.

Feng

, et al. (2018) Optimal design of wind and solar hybrid power system based on particle swarm optimization. Journal of Zhejiang University of Technology 46(06): 650–655.

20.

Yan

Xiang

Zhang

, et al. (2014) Research on dynamic simulation of micro gas turbine based on Matlab/Simulink. Gas Turbine Technology 27(01): 32–37.

21.

Yao

(2020) Research on capacity optimization of hybrid energy storage with hydrogen storage in a wind-solar hybrid power system. Electrical Switch 58(06): 18–22.

22.

Zheng Li, Liping Zhang, Lei Du, et al. (2020) Grid-connected Control Strategy for Controllable Hybrid Systems Based on Hydrogen Storage. Recent Advnces in Electrical & Electronic Engineering 4: 580–594.

23.

Zhao ZZ, Wang WQ, Fan XC, et al. (2021) Multi-objective optimization operation model of microgrid based on NSGA-II-PSO algorithm. Journal of Power Supply 1–14. Available at: https://kns.cnki.net/kcms/detail/12.1420.TM.20201209.1404.004.html

24.

Zhong

Yan

(2019) Research on modeling and Simulink of a micro gas turbine based on Matlab. Power Equipment 33(05): 325–330.

25.

Zhu

Liu

(2017) GA-PSO algorithm based day-ahead economic schedule for islated microgrid. Modern Power 34(01): 15–22.