Multi-objective optimization scheduling for new energy power system considering energy storage participation

YUN Yun-yun, DONG Hai-ying,2, CHEN Zhao, HUANG Rong, DING Kun

(1. School of Automation and Electrical Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China；2. School of New Energy and Power Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China；3. Electric Power Research Institute of State Grid Gansu Electric Power Company, Lanzhou 730070, China)

Abstract： For the low utilization rate of photovoltaic power generation, taking a new energy power system constisting of concentrating solar power (CSP), photovoltaic power (PP) and battery energy storage system as an example, a multi-objective optimization scheduling strategy considering energy storage participation is proposed. Firstly, the new energy power system model is established, and the PP scenario generation and reduction frame based on the autoregressive moving average model and Kantorovich-distance is proposed. Then, based on the optimization goal of the system operation cost minimization and the PP output power consumption maximization, the multi-objective optimization scheduling model is established. Finally, the simulation results show that introducing energy storage into the system can effectively reduce the system operation cost and improve the utilization efficiency of PP.

Key words： new energy power system; multi-objective optimization; energy storage participation; operation cost; autoregressive moving average model

0 Introduction

At present, for the installed capacity of new energy, the total amount of quitting the wind power and photovoltaic power (PP) is still high. Although large-scale renewable energy connected to the grid can bring certain economic benefits, its output power has the characteristic of randomness, which makes the problem of new energy consumption increasingly significant[1-3].

In order to improve the level of new energy consumption, many scholars have solved this problem through reasonable scheduling of demand side resources[4-5]and multi-source complementation in the power generation side[6]. In Ref.[7], the errors of wind speed and solar irradiance, were modeled by related probability distribution functions. And then, by using the Latin hyper-cube sampling, the plausible scenarios of renewable generation for day-head energy and reserve scheduling were generated. A two-stage stochastic objective function aiming at minimizing the expected operational cost was implemented. In Ref.[8], electric vehicles were classified by its schedule ability, and a wind/vehicle coordination optimization strategy was proposed to achieve the goal of increasing wind power utilization efficiency. In Ref.[9], in order to reduce the adverse effects of large-scale renewable energy grid-connected power systems, a hydropower-photovoltaic-wind energy hybrid power generation system scheduling model was proposed. In Ref.[10], an optimization model to coordinate multiple generating units was proposed to improve the regulation capability to accommodate more wind power in power system. However, most of the current studies separately takes thermal power units and gas turbines as the basic power source, but there are few studies on concentrating solar power(CSP) with small environmental pollution and stable output power.

The CSP plant equipped with a heat storage system can sustain the full operation of the plant about 15 h without irradiation[11-12], which has good scheduling characteristic. Moreover, the CSP unit has the ability to climb quickly. Therefore it has the potential to replace conventional energy as the basic power source. The battery energy storage system (BS) can release a large amount of electric energy quickly in a short time, and can be continuously discharged with small output power for a long time[13]. Therefore, the battery energy storage system can not only fill the power shortage due to the prediction error in a short time, but also improve the consumption space of the new energy through its energy storage characteristics. The above analysis shows that the battery energy storage system and the CSP plant have good complementarity.

In order to improve the ability of new energy consumption in Northwest China, this paper presents a multi-objective optimization scheduling strategy for new energy power system considering energy storage participation. Firstly, the model of new energy power system is established. Then, the scenario generation and reduction method based on auto autoregressive moving average model and K-distance is used to simulate the uncertainty of PP output power. Based on this, a multi-objective optimization scheduling model that considers operational economy and new energy consumption level is established. Finally, the simulation shows the economic benefit and effectiveness of the model.

1 New energy power system model

1.1 BS output model

The battery energy storage system has the characteristics of quickly charge and discharge, and its operating cost can be expressed as

(1)

The BS needs to meet the following constraints as

(2)

(3)

(4)

1.2 CSP output model

Because CSP plant is usually equipped with large capacity thermal energy system, the instantaneous irradiance change will not directly affect the output power of CSP plant. Moreover, the change of irradiation in next day can be predicted very accurately[11]. The overall solar energy available in a day can be predicted well[14], meaning that the recovery of storage level will not be bothered by remarkable uncertainty. Therefore, for simplicity, the uncertainty can be ignored while calculating the intervals for joint power output. The predicted solar series can be directly used.

The CSP operating cost can be expressed as

(5)

(6)

(7)

The CSP needs to meet the following constraints as

(8)

(9)

(10)

(11)

1.3 PP output model

Based on the photovoltaic effect, PP converts solar energy into electric power, and the output power is mainly affected by factors such as irradiation intensity and photoelectric conversion rate.

Irradiation intensity is usually described by the Beta distribution function as

(12)

(13)

(14)

whererandrmaxare the irradiation intensity and the maximum irradiation intensity, respectively;αpandβpare the shape distribution parameters of Beta; μp and σp are the average values of the irradiation intensity and standard deviation of irradiation intensity, respectively.

As can be seen from the above, the PP output is

Pt,PP=SPPηPPrt,

(15)

whereηPPis the photoelectric transformation efficiency,SPPis the area of photovoltaic array, andrtis the irradiation intensity at timet.

From Eqs.(1)-(4), the PP output power can be described by the Beta distribution function as

(16)

The output power of PP is mainly affected by the change of irradiance, therefore the decision-maker needs to obtain a representative scenario by scenario generation and reduction method to formulate a scheduling plan.

2 Scenario generation and reduction

Due to the uncertainty of new energy generation, it is very important to draw up the day-ahead dispatching scheduling based on the day-ahead power prediction[15]. In our work, the scenario simulation method is used to describe the uncertainty of photovoltaic power generation, including scenario generation and scenario reduction.

The generated output power scenarios of the photovoltaic power generation by autoregressive moving average model is expressed as[16]

(17)

whereytis the time series value of the periodt;φiis the auto autoregressive parameter;μjis moving average parameter;αtis the normal white noise process in which the average value is 0 and the variance value isσ2.

With the number of scenes increasing, the process of solving the multi-objective scheduling model becomes more and more tedious. Therefore, in order to reduce the computational complexity, we use the K-distance based pushback reduction method to reduce the amount of scenarios.

Supposing thatSandS′ are two large-scale scenario sets, the K-distanceL(S,S′) can be expressed as

(18)

wheresands′ are any scenario in setSand setS′, respectively;δsandδs′are the occurrence probabilities of scenariosand scenarios′ in the scenario setSand scenario setS′, respectively;λ(s,s′) is the distance function;ε(s,s′) is the product of occurrence probabilities of scenariosand scenarios′.

Based on the output power characteristics of photovoltaic power generation, it is assumed that the original photovoltaic output scenario set isSand the target photovoltaic output scenario set isS′. The above formula can be equivalent to

(19)

(20)

The purpose of reducing the scenarios can be achieved by repeatedly calculating Eq.(19).

3 Multi-objective optimization scheduling model of new energy power system

3.1 Objective function

In order to ensure the minimum operation cost of the system and the maximum utilization of PP, the objective function is established as

F1=min(Cyw+CBS+Ccf),

(21)

(22)

(23)

(24)

3.2 Constraint conditions

1) The power balance constraint

(25)

2) System spinning reserve constraints

(26)

(27)

3) PP operation constraint

(28)

The constraints also include BS operation constraints and CSP operation constraints.

3.3 Solution of multi-objective weight coefficient

It can be seen from the above optimization scheduling model that the model has two objective functions, which can be transformed into a single target by setting reasonable weight coefficients. In this paper, the weight coefficients of the optimization target are solved by rough set theory. Relevant knowledge is detailed in Ref.[18]. The solution process is as follows.

1) Relational data model

Assuming that the target weight coefficientfnof the arbitrary objective function is 1/nand the best value of multiple objective functions isF, a decision attribute set with the expressionD={F} can be obtained by usingFas the decision attribute. At this point, there exists a comprehensive optimal valueumin the set, andum=(f1m,f2m,…,fim;Fm). The sample set can be expressed asU={u1,u2,…,um} and the attributes ofumarefn(um)=anm,Fn(um)=Fm.

2) Dependence ofRAonRD

(29)

whereRAandRDare knowledge bases; [F]RDis the knowledge baseRDwhich contains one value ofF;RA(·) is knowledge of the sample setUbefore deleting the indicatoran;d(·) is the rough set base.

3) Dependence ofRAonRA-|an|

(30)

whereRA-|an|(·) is knowledge of the sample setUafter deleting the indicatoran.

4) Weight coefficient

σD=δRA(D)-δRA-|an|(D),

(31)

(32)

whereσDis the importance of the goal, andθnis the weight coefficient of the objective function. Ifa1anda2are assumed to be the weighting coefficients of the minimum operation costs of the system and the maximum accommodation of PP respectively, the sum ofa1anda2is 1.

4 Simulation

4.1 Basic data

The new energy power system consists of PP plant, CSP plant and BS. The installed capacity of PP is 200 MW. After generating and reducing the scenario sets, five typical PV output scenarios are obtained. Taking the average of each scenario as the PP forecast output, the load power prediction result and the photovoltaic output power prediction result are shown in Figs.1 and 2, respectively.

Fig.1 Prediction result of load

Fig.2 Prediction result of PP output power

The maximum and minimum outputs of CSP are 100 MW and 10 MW, respectively. The maximum and the minimum storage levels of the thermal energy storage system are 1 000 MW and 100 MW, respectively. Other parameters can be found in Ref.[17]. The BS has a total capacity of 50 MW and its maximum charge and discharge power are 8 MW. The Light irradiation intensity is shown in Fig.3.

Fig.3 Light irradiation intensity

Since photovoltaic power generation can only generate electricity power during the daytime, simulation analysis only from 7:00 to 18:00 is performed. The maintenance cost of photovoltaic power generation is 30 RMB yuan/(MW·h). The maintenance costs of solar energy collector system and thermal energy system are 50 RMB yuan/(MW·h) and 40 RMB yuan/(MW·h), respectively. The penalty cost of quitting PP is 100 RMB yuan/(MW·h). The operation and maintenance cost and depreciation cost of BS are 25 RMB yuan/(MW·h) and 25 RMB yuan/(MW·h), respectively.

According to Section 3.3, the weight coefficient of two objective functions can be calculated. The results of the minimum operation cost of the system and the maximum utilization of new energy are 0.69 and 0.31, respectively. The prediction error of photovoltaic power generation is 0.05. The model is solved by the Matlab software using the CPLEX 11.0. The solution steps are as follows.

The first step is the problem preprocessing, which means to set the solution accuracy and feasible solution test methods.

The second step is to relax the 0-1 discrete variable of the unit state and solve the relaxation problem. If there is no feasible solution, the software continues to search new nodes. If there exists a feasible solution, the software determines the feasibility of solution by judging whether the solution satisfies the cutting equation. Then, the optimal solution which satisfies the equation is selected as the new lower boundary value.

The third step is to solve the integer variable problem. The relaxed optimal solution is taken as the initial point of integer segmentation optimization. If the integer condition is not feasible, return to the second step to continue the solution. If the condition is satisfied, the upper boundary value of the problem is modified according to the objective function.

The fourth step is to judge the convergence condition. If the relaxation optimal solution and the integer optimal feasible solution satisfy the convergence condition, the calculation is ended. Otherwise, return to the second step to search for a solution that meets the requirements.

4.2 Result analysis

In order to verify the economic benefit and effectiveness of the model, two scenarios are set up for simulation analysis.

Scenario 1: The basic scenario. This scenario includes the PP plant and the CSP plant.

Scenario 2: BS scenario. This scenario includes the PP plant ,BS and CSP plant.

The scheduling result of scenario 1 is shown in Fig.4.

Fig.4 Scheduling result of scenario 1

According to Fig.4, the PP output power and the CSP output power in scenario 1 are 343.7 MW·h and 516.6 MW·h, respectively. The power value of quitting PP is 108.1 MW·h. The consumption rate of photovoltaic power generation is only 76.07%. In this scenario, the CSP plant not only needs to supply electricity power to the load, but also needs to provide reserve capacity for photovoltaic power generation to prevent emergencies. In order to reduce the risk of power shortage, the system increases the power supply proportion of CSP. Although this behavior can improve the reliability of NEPS, it fails to fully exploit the economic benefits of photovoltaic power generation. The scheduling result of scenario 2 is shown in Fig.5.

Fig.5 Scheduling result of scenario 2

According to Fig.5, the PP output power and the CSP output power in scenario 2 are 411.4 MW·h and 469.9 MW·h, respectively. The power value of quitting PP is only 40.4 MW·h. The consumption rate of photovoltaic power generation is 91.06%. The energy storage system has two characteristics of charging and discharging. During the trough period of load power consumption and peak period of PP, the BS stores part of the electric energy and discharges the electricity power during the peak period of load power consumption and the trough period of PP, respectively. Moreover, the BS can provide reserve capacity for PP, improve the consumption space of PP, and reduce the power-supply pressure of CSP. In this way, CSP can provide more reserve capacity, which can improve scheduling flexibility of the dispatching department and reduce the risk of power shortage in the system. The power curves of the thermal energy storage in different scenarios and the CSP output curves in different scenarios are shown in Figs.6 and 7, respectively.

Fig.7 CSP output curves in different scenarios

According to Figs.6 and 7, the BS can reduce the power-supply pressure of CSP and increase thermal energy stored in the thermal energy system. The electricity power space generated by CSP can be supplied by PP, which improves the utilization rate of PP. The PP output curves in different scenarios are shown in Fig.8.

According to Fig.8, the introduction of BS makes grid-connected power value of PP in each period. Moreover, the proportion of PP increases from 39.95% to 46.92%. The comparison of quitting PP in different scenarios and comparison of the system operation costs in different scenarios are shown in Figs.9 and 10, respectively.

Fig.8 PP output curves in different scenarios

Fig.9 Comparison of quitting photovoltaic powers in different scenarios

Fig.10 Comparison of system operation costs in different scenarios

According to Figs.9 and 10, the introduction of BS can reduce the system operation cost and improve the utilization rate of PP. The comparison of the scheduling results in different scenarios is shown in Table 1.

According to Table 1, the system operating cost of scenario 2 is 1 933 RMB yuan lower than that of scenario 1, decreased by about 2.26%. The power value of quitting PP is reduced by 67.7 MW with an decrease of about 62.63%. The utilization rate of PP is increased from 76.07% to 91.06% with an increase of about 15%. The above analysis shows that the model has good economic benefit and improves the consumption rate of PP.

Table 1 Comparison of scheduling results in different scenarios

5 Conclusions

Considering the minimum system operating cost and the maximum PP consumption, the new energy power system multi-objective optimization scheduling problem is studied, and the effectiveness of the model is verified in the simulation. The following conclusions are obtained.

1) When the system contains BS, the power value of quitting PP is significantly reduced, and the output of PP reaches 46.92% of the load demand power. The BS effectively reduces the output power of CSP and improves the grid-connected space of PP. Moreover, the BS stores part of the electrical energy, which improves the scheduling flexibility of the dispatching department. Therefore, BS can effectively improve the consumption rate of PP and reduce the cost of quitting photovoltaic power.

2) The operating cost of new energy power system with BS is 83 715 RMB yuan, while the operating cost of new energy power system without BS is 85 548 RMB yuan. It can be seen that the introduction of BS can reduce the operating cost of new energy power system and improve economic benefits.