# Distributed energy management for interconnected operation of combined heat and power-based microgrids with demand response

- 2.2k Downloads
- 19 Citations

## Abstract

From the perspective of transactive energy, the energy trading among interconnected microgrids (MGs) is promising to improve the economy and reliability of system operations. In this paper, a distributed energy management method for interconnected operations of combined heat and power (CHP)-based MGs with demand response (DR) is proposed. First, the system model of operational cost including CHP, DR, renewable distributed sources, and diesel generation is introduced, where the DR is modeled as a virtual generation unit. Second, the optimal scheduling model is decentralized as several distributed scheduling models in accordance with the number of associated MGs. Moreover, a distributed iterative algorithm based on subgradient with dynamic search direction is proposed. During the iterative process, the information exchange between neighboring MGs is limited to Lagrange multipliers and expected purchasing energy. Finally, numerical results are given for an interconnected MGs system consisting of three MGs, and the effectiveness of the proposed method is verified.

## Keywords

Interconnected microgrids Energy management Distributed optimization Demand response Combined heat and power (CHP)## 1 Introduction

Microgrids (MGs) are self-controlled entities which facilitate the penetration of renewable energy and distributed energy resources (DERs) for economic and reliability purposes. Generally, the MG can be operated in either the grid-connected or islanded mode [1]. With the development of MGs, a new concept of interconnected microgrids system (IMS) (or microgrid cluster) is introduced which considers several MGs exchanging energy with each other even when the MGs are isolated from the utility grid. By constituting the IMS, it is more flexible to ensure the full utilization of renewable energy sources (RESs), reduce the operation cost, and achieve high power supply reliability [2, 3, 4]. From the viewpoint of Transactive Energy, the MGs can be seen as prosumers with both attributes of sellers and buyers. During different time periods, the MG may act as a seller or buyer depending on real-time operating conditions and the net power profile. Therefore, in order to achieve the operation goal of IMS, the energy management is an important issue that should be addressed.

More recently, there were some studies focusing on the energy management of IMS, and the proposed method can be classified into two types: centralized optimization and distributed optimation. Generally, if all the MGs could share the information on their respective data on load, generation, and grid conditions, the optimal scheduling could be easily implemented based on the traditional centralized optimization, such as the optimal power flow (OPF). For instance, a method of joint optimization and distributed control for IMS was proposed in [5], which uses the minimum generation cost as the objective function. However, for security considerations, it is not desirable for each MG to do so because the shared information could compromise the privacy of each MG. Thus, this is the basic motivation for the deployment of distributed optimization. In this regard, more attention has been paid to the distributed optimizations for IMS energy management. A decentralized optimal control algorithm for distribution management systems was proposed in [6] by considering distribution network as coupled microgrids. The optimal control problem of IMS is modeled as a decentralized partially observable Markov decision process, which decreases the operating cost of distributed generation and improves the efficiency of distributed storages. Moreover, the alternating direction method of multipliers (ADMM) was applied in [7] and [8] for optimal generation scheduling of IMS. Only the expected exchanging power information needs to be shared among all the MGs during the iterative process to minimize the total operation cost. Similarily, a distributed convex optimization framework is developed for energy trading among islanded MGs in [9] and [10] with the objective of minimizing the total operation cost.

- 1)
Considering the power and heat demands and the possible energy trading among MGs, an hour-ahead optimal scheduling model is proposed. The system model considers the cost of DERs, the cost of DR, the network tariff, and the power loss of interconnected power lines.

- 2)
A distributed iterative algorithm based on subgradient with dynamic search direction is proposed, in which the search direction is constructed by combining conjugacy and subgradient method.

## 2 System model

### 2.1 Distributed energy resource

*i*.

*j*; and \(H_{chpj}\) is heat generation of CHP

*j*.

*k*.

### 2.2 Demand response

### 2.3 Network cost

*a*and

*b*are coefficients; and

*x*is the trading energy.

### 2.4 Optimal scheduling model

Consider an IMS consisting of *M* interconnected MGs through a power interconnection infrastructure and a communication network. Let \(E_i^{(g)}\) and \(E_i^{(c)}\) be the generation and consumption of MG *i* during each scheduling time slot, respectively. MG *i* is allowed to sell energy \(E_{i,j}(E_{i,j}\ge 0)\) to MG *j*, \(j\ne i\), and can buy energy \(E_{k,i}(E_{k,i}\ge 0)\) from MG *k*, \(k\ne i\). In order to describe the connection between MGs, an adjacency matrix \({{\varvec{A}}}=[a_{i,j}]_{M\times M}\) is defined. If there exits a connection from MG *i* to MG *j*, element \(a_{i,j}\) is set as 1; otherwise, element \(a_{i,j}\) is set to be 0. Thus, \({{\varvec{A}}}\) may be nonsymmetric, meaning that at least two MGs are allowed to exchange energy in one direction only. Moreover, we choose \(a_{i,i}=0\), and if \(a_{i,j}=0\), we can directly have \(E_{i,j}=0\).

*i*; \(\gamma (E_{i,j})\) is the cost of transferring \(E_{i,j}\) units of energy between MG

*i*and MG

*j*; \(\mathbf e _i\) is the

*i*th column of the \(M\times M\) identity matrix; \({{\varvec{E}}}_i^{(b)}\) is the vector composed of the energy bought from other MGs by MG

*i*; \(\varvec{\gamma }({{\varvec{E}}}_i^{(b)})=[\gamma (E_{1,i})\cdots \gamma (E_{M,i})]^{\rm T}\).

The coupled multiple MGs in one IMS, which have their set of possible actions, should be coordinated in order to achieve the common goal of the system and meet the power and heat demands.

*i*, its total operation cost \(C_i(E_i^{(g)})\) includes the cost of DG, CHP, heat-only unit and virtual generation unit.

*i*; \(N_{chp}\) is the number of CHPs in MG

*i*; \(N_h\) is the number of heat-only units in MG

*i*; \(C_{dgj}\) is the cost function of DG

*j*; \(C_{chpk}\) is the cost function of CHP

*k*; \(C_{hm}\) is the cost function of heat-only unit

*m*; and \(C_{DRi}\) is the cost function of the virtual generation unit of DR in MG

*i*.

The optimal scheduling problem has several constraints, which can be classified as follows.

1) Power balance

*i*requires:

*j*by MG

*i*;

*U*is the voltage of interconnection lines; and \(R_{ij}\) is the resistance of the interconnection line between MG

*i*and MG

*j*.

Moreover, any power transfer between MGs is accompanied with a cost of power loss over the interconnection lines. We assume that the reactive power is compensated for by each MG individually. Also it is assumed here that the cost of the power loss between MGs is covered by the power purchaser.

2) DR constraint

*v*is the pre-specified portion of the nominal load, and thus (17) guarantees that the load curtailment is smaller than a pre-specified portion of the nominal load.

3) Power constraints

*i*.

## 3 Distributed model and algorithm

### 3.1 Distributed optimal scheduling model

Problem (10) is known to have a unique minimum point since both the objective function and the constraints are strictly convex. As discussed in Sect. 1, there are difficulties for centralized optimization applied to IMS. In this regard, we decide to propose a distributed optimal scheduling model by decomposing the problem (10) into *M* local subproblems, which can be implemented by the MGs in an autonoums and cooperative manner.

In the above equations, \(\varepsilon _i^{(s)}\) denotes the total selling energy of MG *i*, which is forced to be equal to all the energy bought by other MGs from MG *i*. A coupling constraint is formed as follows: \(\varepsilon _i^{(s)}={{\varvec{e}}}_i^{\rm T}{{\varvec{A}}}{{\varvec{E}}}_i^{(s)}\).

*i*to the Lagrangian function relative to (10). Based on the above analysis, each Lagrange multiplier \(\lambda _i\) can be interpreted as the marginal cost of MG

*i*, namely the price that selling a unit of power to adjacent MGs. Thus Lagrange function (28) can be seen as the net expenditure. The expenditure of each MG consists of the following parts: \(\textcircled {1} \,C_i(E_i^{(g)})\) is the generating cost including various generation units; \(\textcircled {2}\) \({{\varvec{e}}}_i^{\rm T}{{\varvec{A}}}^{\rm T}\varvec{\gamma }({{\varvec{E}}}_i^{(b)})\) is the network cost resulted from transferring the energy purchased from other MGs; \(\textcircled {3}\) \({{\varvec{e}}}_i^{\rm T}{{\varvec{A}}}^{\rm T}{\rm diag}{\{\varvec{\lambda }\}}{{\varvec{\textit{E}}}}_i^{(b)}\) is the cost due to purchasing energy; and \(\textcircled {4}\) \(\lambda _i\varepsilon _i^{(s)}\) is the income by selling energy.

### 3.2 Distributed algorithm

Obviously, the problem is transformed to the maximum dual problem. To this end, the optimal Lagrangian multipliers which converge to the optimal point of the dual problem are necessary to be found, \(\varvec{\lambda }^*={\rm argmax}_{\varvec{\lambda }}C(\varvec{\lambda })\). For each point \(\varvec{\lambda }[k]\), each MG minimizes its contribution to the Lagrangian function by solving the local subproblem (27) and determining the minimum point. As subproblem (27) is a convex function, we use interior point method to obtain the optimal solution.

According to [19], the conjugate gradient method is used to solve for the minimum value of the function, which has the quadratic termination property. Combining conjugacy and subgradient method, it shows a better convergence performance. In the conjugate gradient method, the search direction is constructed by taking *n* steps as a round and taking the negative gradient direction for the initial search direction of each round. Thus, referring to the conjugate gradient method, a subgradient method considering the dynamic search direction is developed. In this paper, we aim to search for the maximum value of the dual problem (26). Therefore, during the iteration, the initial search direction of each round is a subgradient direction.

The subgradient of \(C(\varvec{\lambda })\) in \(\varvec{\lambda }=\varvec{\lambda }[k]\) can be described as \(\varvec{\varsigma }=[{{\varvec{e}}}_i^{\rm T}{{\varvec{A}}}{{\varvec{E}}}_i^{(s)}[k]-\varepsilon _i^{(s)}[k]]_{M\times 1}\). For \(\forall \varvec{\lambda }\), we have \(C(\varvec{\lambda })\le C(\varvec{\lambda }[k])+\varvec{\varsigma }_T(\varvec{\lambda }-\varvec{\lambda }[k])\).

*n*steps as a round, and the initial update function of the Lagrange multipliers in each round can be expressed as:

*k*is the iteration number.

Next, when the convergence condition is not satisfied and *m* ( *m* is the iteration variable of determining the search direction) is no longer less than *n*, we take the next round according to (29), (30), (31) and (32).

Algorithm 1 summarizes the steps of the proposed distributed iterative algorithm.

Having solved (27) in all MGs, each MG can be aware of \(\varepsilon _i^{(s)}[k]\) and \({{\varvec{E}}}_i^{(b)}[k]\), namely the total energy it sold and the vector composed of the energy bought from other MGs. Furthermore, we can obtain \({{\varvec{E}}}_i^{(s)}\) from \({{\varvec{E}}}_i^{(b)}\) according to (14). Combined with Algorithm 1, the Lagrangian multipliers can be updated. Therefore, all data we need can be calculated by each MG without a centralized controller. In addition, the information exchange between MGs is limited to Lagrange multipliers \({\lambda _i}\) and the expected purchasing energy \({E_{j,i}}\), which is only communicated to the corresponding MG *j*. Therefore, the privacy of MGs can be preserved.

According to Algorithm 1, the price \(\lambda _i\) would be modified constantly before the supply-demand balance. When the energy offered by MG is less than the requested energy from other MGs, the price will be increased as the demand exceeds supply; whereas the price will be decreased as the demand is less than supply. The price remains constant when the supply matches the demand.

## 4 Numerical results

### 4.1 Basic data

Fuel coefficients and capacity of DGs

Units | \(\alpha _i\) | \(\beta _i\) | \(\gamma _i\) | \(P_{G1}^{{\rm min}}\) | \(P_{G1}^{{\rm max}}\) |
---|---|---|---|---|---|

(MWh) | (MWh) | ||||

DG1 | 10.193 | 210.36 | 250.2 | 0 | 0.5 |

DG2 | 2.305 | 301.4 | 1100 | 0.04 | 0.2 |

Fuel coefficients of CHP and heat-only units

Units | \(\alpha _j\) | \(\beta _j\) | \(\gamma _j\) | \(\delta _j\) | \(\theta _j\) | \(\zeta _j\) |
---|---|---|---|---|---|---|

CHP1 | 339.5 | 185.7 | 44.2 | 53.8 | 38.4 | 40 |

CHP2 | 100 | 288 | 34.5 | 21.6 | 21.6 | 8.8 |

Heat1 | 33 | 12.3 | 6.9 |

Capacity of CHP and heat-only units

Units | \(P_{G1}^{{\rm min}}\) | \(P_{G1}^{{\rm max}}\) | \(P_{G1}^{{\rm min}}\) | \(P_{G1}^{{\rm max}}\) |
---|---|---|---|---|

(MWh) | (MWh) | (MWh) | (MWh) | |

CHP1 | 0.05 | 1 | 0 | 0.6 |

CHP2 | 0.05 | 0.6 | 0 | 0.6 |

Heat1 | 0 | 2 |

Demand versus price coefficients

Coefficients | MG1 | MG2 | MG3 |
---|---|---|---|

\(a^{lin}\) | 1 | 1 | 1 |

\(b^{lin}\) | −0.002 | −0.001 | −0.0035 |

### 4.2 Results and analysis of distributed optimal scheduling

1) Trading prices

2) Trading energy

The energy trading after convergence in the current time slot can be explained as follows: MG1 purchases 0.1675 MWh energy from MG3 including \(8.4\times 10^{-4}\) MWh as power loss; MG2 purchases 0.1587 MWh energy from MG3 including \(8\times 10^{-4}\) MWh power loss; MG3 sells 0.3261 MWh. As we can observe, the total energy sold is equal to the total energy bought in the IMS. The coupling constraint \(\varepsilon _i^{(s)}={{\varvec{e}}}_i^{\rm T}{{\varvec{A}}}{{\varvec{E}}}_i^{(s)}\) is satisfied after convergence, which proves that the algorithm performs well.

During the optimization, the cost of power loss caused by power transmission between MGs is covered by the energy buyer. In this regard, the power loss is also taken into consideration during the distributed optimal scheduling.

In this time slot, MG1 purchases energy from MG3 to meet its load demand, as the marginal cost of its own generation unit is higher than the sum of selling price and the network cost of MG3. Similarly, the marginal cost of DG2 in MG2 is not economical, thus it is better to work on the lower generation limit. The insufficient load demand of MG2 is supplied by the generation of CHP2, curtailing load through DR and purchasing power from MG3.

3) DR

By calculation, the ratios of curtailed load in MG1, MG2 and MG3 are 16.96\(\%\), 7.88\(\%\) and 21.73\(\%\), respectively. Compared to Table 4, the curtailed loads in different MGs have direct relationships with coefficients \(|b_{lin}|\). For example, the load demand of MG3 is most sensitive to the DR incentive, thus the ratio of curtailed load is much higher than other MGs. Moreover, The total costs with DR and without consideration DR are 924.6475 $ and 935.0376 $ respectively. The total operation cost can be reduced through DR under the premise of meeting the basic load demand of each MG.

4) Iterative process of variables

All optimal variables including the selling energy, buying energy, generation, and curtailed load can be solved by Algorithm 1. Taking MG1 as an example, Fig. 7 shows the iterative processes of variables in the decetralized model of MG1.

Comparison with several related papers

Properties | Reference [5] | Reference [7] | Reference [9] | This paper |
---|---|---|---|---|

Exchanged information | All data of sources and load transmitted to control center | Expected exchange power sharing among all the MGs | Price and expected purchasing energy with neighboring MG | Price and expected purchasing energy with neighboring MG |

Distributed generation | DG | DG PV WT | DG | DG CHP PV WT |

Consider DR | No | No | No | Yes |

Consider power loss | No | No | No | Yes |

The number of MGs | Multiple | Two | Multiple | Multiple |

Solution algorithm | Centralized optimization | ADMM | Based on subgradient | Based on subgradient considering dynamic search direction |

Iteration number | Without iteration | 49 | 74 | 38 |

Iteration time (s) | − | 5.4526 | 8.4147 | 4.1367 |

5) Benefits of interconnection

The results show that trading not only reduces the total operation cost, but also cuts down the expenditure of each individual MG. This is because MG3 achieves revenue by selling energy whereas MG1 and MG2 decrease their cost by purchasing energy.

### 4.3 Comparison with the related work

In order to illustrate the benefits and advantages of the proposed model and algorithm, the results are compared to several related papers mentioned in the Introduction section, in terms of exchanged information, the type of DERs, DR, power loss, the number of MGs, solution algorithm and performances. The comparative results are shown in Table 6 where algorithm performance indicators including iteration number and iteration time are obtained based on the same test case. Note that the method in [7] can only be applied for two interconnected MGs. Thus, we only use part of the IMS (as shown in Fig. 1, MG1 and MG2) as the test case for method of [7].

Cost comparison between centralized optimization and distributed optimization of IMS

MG | Cost($) | |
---|---|---|

Centralized | Distributed | |

optimization | optimization | |

1 | 183.1454 | 183.1455 |

2 | 343.1793 | 343.1794 |

3 | 398.3498 | 398.3496 |

Total | 924.6745 | 924.6745 |

As for the exchanged information, [5] which belongs to the centralized optimization requires all measured data of sources and load to be transmitted to the system control center, which results in more requirements on the overall communication cost. Besides, sharing information of load and sources can lead to serious privacy and business information leakage, since MGs may belong to different business owners. For [7], all the expected exchange power of MGs should be shared with each other in the IMS. In this paper, the method is developed based on the distributed optimization framework of [9], the information exchanged among MGs is limited to Lagrange multipliers and the expected purchasing energy quantities, which are only communicated with the trading MGs.

The initial prices of MG1 in this paper and [9] are same before iteration. In this paper, the search direction of first iteration is the subgradient direction, which is same as initial search direction in [9]. Therefore, the prices of MG1 in this paper and [9] are same at the first iteration. Next, the algorithm based on subgradient with dynamic search direction has a faster iteration speed. Finally, the prices of MG1 in this paper and [9] converge to the same value. Obviously, the proposed algorithm has a better convergence performance. Considering that the MGs should be operated in a distributed manner, better convergence speed would finally lower the interaction time with less data exchanges.

Having gained insight into this result, the search routine of subgradient algorithm seems sawtooth shaped. In the local space, the subgradient is the fastest direction for the increasing of objective function value. Thereby, it should be a good choice to search on the subgradient direction. However, in the global space, the convergence speed would be slowed down due to the existence of sawtooth shaped routine. For this drawback, we have extended the subgradient algorithm with the dynamic search directions. During each round of iteration, the initial search direction is obtained by subgradient; after that, the following search directions are constructed based on the combination of conjugacy and subgradient methods. By using the dynamic search direction, the proposed algorithm has addressed the problem caused by the searching routine of sawtooth type, and eventually expedited the convergence.

## 5 Conclusion

In this paper, we present a distributed energy management method for interconnected operation of CHP-based MGs. An hour-ahead optimal scheduling model is built, and the objective function includes the operation cost of CHPs, DGs, DR and network tariff. Considering each MG is operated independently, the optimal scheduling problem is decentralized into *n* sub-problems in accordance with the number of the associated MGs. Moreover, a distributed iterative algorithm is proposed based on the subgradient method considering the dynamic search direction. From numerical simulations, we have shown that each MG can choose to curtail load, adjust generation of DGs or trade with other MGs with a comprehensive consideration of generation cost, trading price, load characteristic and DR cost, which eventually reduces operation costs and makes power utilization more flexible and more interactive. Compared with the related studies, we have also shown the advantageous features in the proposed method on modeling and algorithm performance.

## Notes

### Acknowledgements

This work was supported by the the National High Technology Research and Development Program of China (863 Program) (No. 2014AA052001), and the Fundamental Research Funds for the Central Universities (No. 2015ZD02).

## References

- [1]Oureilidis KO, Bakirtzis EA, Demoulias CS (2016) Frequency-based control of islanded microgrid with renewable energy sources and energy storage. J Mod Power Syst Clean Energy 4(1):54–62. doi: 10.1007/s40565-015-0178-z CrossRefGoogle Scholar
- [2]Resende FO, Gil NJ, Lopes JAP (2011) Service restoration on distribution systems using multi-microgrids. Eur Trans Electr Power 21(2):1327–1342CrossRefGoogle Scholar
- [3]Shahnia F, Chandrasena RPS, Rajakaruna S et al (2014) Primary control level of parallel distributed energy resources converters in system of multiple interconnected autonomous microgrids within self-healing networks. IET Gener Transm Distrib 8(2):203–222CrossRefGoogle Scholar
- [4]Nutkani IU, Loh PC, Blaabjerg F (2013) Distributed operation of interlinked AC microgrids with dynamic active and reactive power tuning. IEEE Trans Ind Appl 49(5):2188–2196CrossRefGoogle Scholar
- [5]Li Y, Liu N, Zhang J (2015) Jointly optimization and distributed control for interconnected operation of autonomous microgrids. In: Smart grid technologies—Asia (ISGT ASIA), 2015 IEEE Innovative, Bangkok, pp. 1–6Google Scholar
- [6]Wu J, Guan X (2013) Coordinated multi-microgrids optimal control algorithm for smart distribution management system. IEEE Trans Smart Grid 4(4):2174–2181CrossRefGoogle Scholar
- [7]Li Y, Liu N, Wu C et al (2015) Distributed optimization for generation scheduling of interconnected microgrids. In: Power and energy engineering conference (APPEEC), 2015 IEEE PES Asia-Pacific, Brisbane, QLD, 2015, pp. 1–5Google Scholar
- [8]He M, Giesselmann M (2015) Reliability-constrained self-organization and energy management towards a resilient microgrid cluster. In: Innovative smart grid technologies conference (ISGT), 2015 IEEE Power and Energy Society, Washington, DC, pp. 1–5Google Scholar
- [9]Gregoratti D, Matamoros J (2013) Distributed energy trading: the multiple-microgrid case. IEEE Trans Ind Electr 62(4):2551–2559CrossRefGoogle Scholar
- [10]Fathi M, Bevrani H (2013) Statistical cooperative power dispatching in interconnected microgrids. IEEE Trans Sustain Energy 4(3):586–593CrossRefGoogle Scholar
- [11]Liu N, Yu X, Wang C et al (2017) An energy sharing model with price-based demand response for microgrids of peer-to-peer prosumers. IEEE Trans Power Syst. doi: 10.1109/TPWRS.2017.2649558
- [12]Ma L, Liu N, Zhang J et al (2016) Energy management for joint operation of CHP and PV prosumers inside a grid-connected microgrid: a game theoretic approach. IEEE Transactions on Industrial Informatics 12(5):1930–1942Google Scholar
- [13]Liu N, Chen Q, Liu J et al (2015) A heuristic operation strategy for commercial building microgrids containing EVs and PV system. IEEE Trans Ind Electr 62(4):2560–2570CrossRefGoogle Scholar
- [14]Aghaei J, Alizadeh MI (2013) Multi-objective self-scheduling of CHP (combined heat and power)-based microgrids considering demand response programs and ESSS (energy storage systems). Energy 55(1):1044–1054CrossRefGoogle Scholar
- [15]Piperagkas GS, Anastasiadis AG, Hatziargyriou ND (2011) Stochastic pso-based heat and power dispatch under environmental constraints incorporating CHP and wind power units. Electr Power Syst Res 81(1):209–218CrossRefGoogle Scholar
- [16]Ma L, Liu N, Wang L et al (2016) Multi-party energy management for smart building cluster with PV systems using automatic demand response. Energy Build 121:11–21Google Scholar
- [17]Yousefi S, Moghaddam MP, Majd VJ (2011) Optimal real time pricing in an agent-based retail market using a comprehensive demand response model. Energy 36(9):5716–5727CrossRefGoogle Scholar
- [18]Saad W, Han Z, Poor HV (2011) Coalitional game theory for cooperative micro-grid distribution networks. In: 2011 IEEE international conference on communications workshops (ICC), Kyoto, pp 1–5Google Scholar
- [19]Chen B (2005) Optimization theory and algorithm. Tsinghua University Press, BeijingGoogle Scholar
- [20]Liu N, Tang Q, Zhang J et al (2014) A hybrid forecasting model with parameter optimization for short-term load forecasting of micro-grids. Appl Energy 129:336–345CrossRefGoogle Scholar

## Copyright information

**Open Access**This 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.