Portal de Periódicos da CAPES

Hierarchical optimisation strategy in microgrid based on the consensus of multi‐agent system

2018; Institution of Engineering and Technology; Volume: 12; Issue: 10 Linguagem: Inglês

10.1049/iet-gtd.2017.0393

1751-8695

Ran Hao, Ziqing Jiang, Qian Ai, Zhiwen Yu, Yuchao Zhu,

Optimal Power Flow Distribution

IET Generation, Transmission & DistributionVolume 12, Issue 10 p. 2444-2451 Research ArticleFree Access Hierarchical optimisation strategy in microgrid based on the consensus of multi-agent system Ran Hao, Ran Hao Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorZiqing Jiang, Ziqing Jiang Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorQian Ai, Corresponding Author Qian Ai aiqian@sjtu.edu.cn Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorZhiwen Yu, Zhiwen Yu Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorYuchao Zhu, Yuchao Zhu Xi'an Jiao Tong University, No.28 Xianning Rd., Beilin District, Xi'an, Shanxi, People's Republic of ChinaSearch for more papers by this author Ran Hao, Ran Hao Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorZiqing Jiang, Ziqing Jiang Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorQian Ai, Corresponding Author Qian Ai aiqian@sjtu.edu.cn Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorZhiwen Yu, Zhiwen Yu Shanghai Jiao Tong University, No.800 Dongchuan Rd., Minhang District, Shanghai, People's Republic of ChinaSearch for more papers by this authorYuchao Zhu, Yuchao Zhu Xi'an Jiao Tong University, No.28 Xianning Rd., Beilin District, Xi'an, Shanxi, People's Republic of ChinaSearch for more papers by this author First published: 28 March 2018 https://doi.org/10.1049/iet-gtd.2017.0393Citations: 6AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract To improve the automation level of distributed generation, a hierarchical optimisation strategy is proposed in this study. The strategy consists of day-ahead dispatch and scheduling implementation by power control. The energy management framework about the multi-agent system is also designed. Given the collaborative gaming process between microgrid and distributed network, a day-ahead dispatch is used to minimise the general expenses. Moreover, considering security constraints, the secondary control strategy is proposed to realise the precise control of the active power, which is adaptive to voltage inconsistency. Besides, the consensus algorithm is utilised to trace the dispatch target of tie-line power by monitoring power deviation at the point of common coupling. Finally, a series of simulation verifies the effectiveness of the method proposed. The influence of communication delay is also discussed. 1 Introduction The sustainable development is facing enormous challenges on how to overcome the problems of energy distribution imbalance and the difference of energy characteristics. Thus, in 2010, future renewable energy transmission and management emphasises the necessity of power electronics and information technology into future power systems [1], which is an effective method to achieve large-scale renewable energy integration in distribution networks. Jeremy Rifkin [2] also pointed out that there was an irresistible trend to make full use of Internet technology to achieve wide-area collaborative control of distributed source, network, load and storage. With the deep integration of the communication network and power network, energy consumption structure and energy utilisation pattern have been undergoing profound changes, which promotes the evolution of regional energy systems to large-scale cyber-physical energy systems [3, 4]. At the same time, with the substantial increase of the heterogeneity of communications and power conversion technologies, microgrid (MG) has been studied as the cyber-physical system in feature analysis [4–6]. Cyber-physical MG system [7–10] combines the cyber components and physical components to obtain common goals such as automatic generation control (AGC) and automatic voltage control. AGC plays a vital role in following the instructions issued by the power dispatch and adjusting real-time power outputs according to a certain adjustment rate. In general, AGC comprises of frequency control, tie-line bias control (TBC) and economic dispatch (ED) [11, 12]. Specifically, ED aims to reduce generation costs through dispatching generation units under various constraints. In contrast, TBC is designed to quickly adjust the exchange power to achieve economical operation of the region. Similarly, to realise the adaptive adjustment, TBC and ED, a hierarchical optimisation strategy is proposed in this paper. In the proposed hierarchical optimisation strategy, ED function of MG is implemented by a centralised control agent in upper layer, while the function of TBC is conducted in communication layer by the multi-agent system (MAS). Droop control is applied extensively in decentralised agents which can be regarded as primary control. As a result, the strategy upgrades the automation level in MG. Specifically, many specific methods can be used to tackle this optimisation problem, one of which in high performance is proportional method. It regulates a fixed participation factor to each generation tribe which is proportional to the adjustable marginal cost of units. Recent advances in animal behaviour, aviation control and traffic dispatch have facilitated the investigation of consensus algorithm. So far, some studies involved power control with collaborative consensus algorithm (CCA) [13, 14], in which agents can communicate with neighbours to reach a consensus through some state variables. However, no previous studies investigated the application of tie-line power control by CCA. In [15], dynamic synergetic MAS framework is proposed and the dispatch algorithm considering security domain is provided, while it ignores the economy of MAS. Majumder et al. [16] builds a multi-objective optimisation model of MG with the active distribution network (ADN) and MGs, but it does not pay much attention to the precise control of dispatch orders. Early studies of the control strategy in MG focus on the improvement of droop control [17, 18], where several improved strategies are put forward to cope with precise power allocation in MG. Most of the current literatures concentrate on the consistency control strategy based on P –F, Q –U droop control [19–21]. However, a minority is concerned with the situation where system's impedance is resistive. So far, to analyse the effect of transmission delay and noise, necessary experiments are provided in [22]. The authors in [23, 24] set the internal boundary price curve of the economic droop control strategy. However, it cannot be applied to the real-time electricity prices and multi-functional collaborative application scenarios. In this paper, MG optimal control strategy is studied. First, for the convenience of agents coordination, a hierarchical framework of MAS is established. Second, combining the operation model of the active distributed network and dispatch model of MG, a bilayer day-ahead dispatch model based on the cooperative evolutionary game model is built to optimise tie-line power. Internal dispatch in MG is realised. Moreover, this model enables ADN and MG to compete for comprehensive benefits, where the total costs will be reduced. Third, the secondary distributed control based on the CCA is designed. Under the premise of safety and stability, active power is controlled accurately by proportion. On this basis, a synergetic strategy of tie-line power traces real-time instructions of dispatch, which advances the automation level of energy management. As a result, the total costs of operation will decrease. Besides, the effect of communication delay is discussed. Finally, simulations and numerical analysis validate robustness and effectiveness of the strategy proposed. This paper is organised as follows. Section 2 discusses the collaborative framework of MG based on MAS; Section 3 presents optimal dispatch model of centralised dispatching centre. Considering the interaction between ADN and MG, the game algorithm is designed to optimise the economic operation. In Section 4, secondary consensus control including active power control and tie-line power control is introduced. In Section 5, the results of synergetic control strategy's simulation based on CCA are illustrated and analysed using a case of five-node MG. Also, the influence of communication delay on consensus algorithm is discussed. 2 MG framework based on MAS for automatic generation Nowadays, the number of generation nodes grows in geometric multiplier and information interaction is increasingly complex and diverse. The traditional centralised optimisation is faced with unprecedented pressure and challenges on the communication response time and computing ability, while the simple distributed solution has the problem of low efficiency and only partial optimal solution. Thus, multi-agents' function of integrating physical and cyber resources is indispensable in achieving components' coordination. MAS where agents are fully interactive can respond intelligently and flexibly to the changes in operation conditions and the needs of the surrounding environment [25]. Therefore, this paper establishes the MAS for MG synergetic control strategy. The proposed MAS in MG consists of a set of agents that are closely related to the physical and information levels. Thus, according to the synergetic strategy proposed in this paper, five kinds of agents are designed, as shown in Fig. 1. The functions of these five kinds of agents are described in detail as follows. Fig. 1Open in figure viewerPowerPoint Hierarchical architecture in MG with automatic generation Management agent : The manager of ADN and MG by controlling tie-line power. Connected with ADN, it can receive the instructional control signal from distribution network and collect real-time information about point of common coupling (PCC) point, which includes the breaking signals and power exchanges. MG centre control agent (MGCCA) : Optimal dispatch agents. It cooperates with the energy hub agent (EHA) to maximise the economic benefit in day-ahead dispatch. To ensure the power balance and voltage stability of MG according to real-time electricity prices, it obtains the dispatching plan by calculating the optimal generation ratio. EHA : Coordinator of primary and secondary energy to meet cooling and heat load. The major functions are: calculation of the available capacity of primary energy, analysis of users' energy demand and coordinating coupling elements (such as gas turbine, cogeneration of heat and power) output with MGCCA. Distributed generation agent (DGA) : Synergetic agents of distributed generation. It can be argued that the energy storage with strong flexibility is set as dominate distribution generation agent (DDGA). DDGA selects operation modes according to the remaining capacity of energy storage. It can also receive initiate values from MGCCA if the fluctuation of tie-line power exceeds a certain threshold. After receiving an initial value, DDGA starts consensus iteration between DGAs immediately. On the other hand, DGAs receive the generation ratio from MGCCA and calculate the economic droop coefficients based on the maximum voltage deviation. Secondary active power control in DGAs ensures accurate power sharing by CCA. Load agent : Static agents monitoring load feeders. In the case of unplanned islands and fault conditions, it is used in conjunction with DGAs to divide non-critical loads in batches. 3 ED of MGCCA This section brings a coordinated ED method of DN and MG. Considering the interaction of MG in ADN operation, MGCCA regards the energy Internet and ADN as different stakeholders. It establishes the benefit model of ADN and MG, respectively. In consideration of the game relationship between different subjects, a co-evolutionary game algorithm (CGA) is designed, which takes ADN and MGs as game participants. 3.1 ADN optimisation model MGs can respond to the regional grid for peak shaving and valley filling. Also, its reserve capacity can be sold to ADN. Here total number of dispatch cycle is defined as T. Regardless of the economy and operation constraints in MG, ADN's optimisation objective is tantamount to minimising general costs including peak shaving and reserve capacity. The specific benefit function is as follows: (1) (2) (3) (4) where is the costs of purchasing reserve capacity; , indicate the unit price of reserve capacity and the capacity at time t, respectively; is the peak shaving price of ADN; is the load at time t ; is average load per day; is the output power of MG at time t ; is the peak-valley difference of load. Reserve capacity constraint (5) where is the redundancy of load reserve capacity. Power balance constraint (6) where represents the active power assigned to at time t ; is the tie-line power at time t. Tie-line power limit (7) , are criticle values of tie-line power. 3.2 Internal ED in MG The optimisation objective of MG is to minimise the overall operating costs, illustrated by (8) Here, , , , , indicate fuel costs, operating maintenance costs, installation depreciation costs of MG, pollution costs and power loss at time t, respectively. represents the revenue provided by ADN. Detailed expressions of each optimisation targets are given in [26–28]. Power generation constraints (9) Unit ramp rate constraints (10) where , are the upper and lower bounds of power generation assigned to ; and are the maximum and minimum ramp rate assigned to . 3.3 Tie-line power game algorithm In order to satisfy the power balance constraint (6), premise control of tie-line is indispensable. This paper designs the synthetic game algorithm of DN and MG, which could simulate the game process of two individuals competing for limited resources. The economic benefits of both DN and MG are regarded as the optimisation target. A fuzzy process is designed making unified numerical values for comparison [29]. Define the membership function u. The size of u reflects the degree of optimisation. u = 1 represents the best degree of optimisation, while u = 0 indicates the worst one. Through comparative analysis, we choose Γ-shaped distribution in this paper (11) In the cooperative evolutionary game algorithm, DN and MG are used as game participants to calculate the tie-line power. Two populations and are defined. Each population has a certain number of individuals, which record not only the decision vector but also the belonging population. and are optimisation targets. The CGA draws on the experience of selection mechanism of the evolutionary game. When two individuals ( ) in the group meet, they play a game for the same resource, and the payment function of this optimisation problem is as follows: (12) where , are the minimum and maximum values of the population, respectively. The payment function indicates the obtained resources after a single game. In each generation of the simulation, pairs of individuals are selected randomly to carry out a number of repeated games. When all the games are completed, the average payment of the individual x for is regarded as fitness (13) The deviation of the tie-line power () generated by the game needs to be limited. Maximum power fluctuation of can be determined by the power fluctuation expense penalty function. The penalty function at time t can be written as and should meet the inequality constraint (14) where is unit penalty cost. is the maximum dissatisfaction and is the integrated cost of MG at t. Finally, the adaptive genetic algorithm is used to solve the game optimisation model. The highest value of benefit function is the optimal power of tie-line. 4 Secondary synergetic control strategy based on CCA Before the strategy is discussed, a brief introduction of CCA is indispensable. Generally, the discrete consensus algorithm [21] can be denoted as (15) Parameter d is determined by a doubly stochastic matrix. Wang and Xiao [30] proposed Metropolis doubly stochastic matrix of the construction method, which can be expressed as (16) where represents the maximum value of node i and its neighbouring nodes. In the connected matrix, represents that node i is adjacent to node j. 4.1 Precise active power control The droop control can implement primary adjustment of voltage and frequency independently by droop characteristics. It still has a drawback of state offset caused by the different operating conditions. This part analyses this issue and gives an improved strategy. In P –V, Q –F droop control, the DG reference voltage and frequency can be expressed as (17) (18) where droop economic coefficients of P –V and Q –F can be expressed as (19) (20) In the formula, and are the reference values of voltage and frequency, respectively; , , and are rated voltage, the maximum voltage and the corresponding frequency, respectively; and are calculated by (25) and (26) according to and ; denotes the reference active power output assigned to given by day-ahead dispatch; is the inherent characteristics of maximum reactive power output assigned to . MGCCA communicates with DGAs once a day to transmit generation scheduling of next day. Line impedance and droop control will produce voltage and frequency offset. In order to prevent large voltage offset, we stipulate , where denotes maximum voltage deviation. When a system is stable, all frequencies in MG are the same, namely . From (20), it is obvious that under the circumstance of the stable operation, the line reactance is small while the reactive power output and reactive droop coefficients are proportional (21) Considering the influence of line resistance, (24) can be deformed into (22) Right side of the above formula can be written as (23) Traditional control method fixes the maximum voltage . However, the terminal voltage of is not equal, so is . As a result, power cannot be accurately assigned (24) Each node voltage is influenced by the line resistance and the line active power. When a feeder is connected to multiple DGs, the active output will be affected by the voltage of adjacent inverters, which results in inaccurate control of the active power and tie-line power. To solve the problem, considering the addition of secondary active control based on CCA, adaptive adjustment is added, where . In this paper, the secondary precision control of active power by adding compensation is designed to conduct the adaptive adjustment of the active power. The DG's offset is the relative average voltage offset (25) The average voltage of the whole network is obtained by synergetic consensus iteration. The discrete consensus iterative formula can be expressed as (26) In this paper, the average voltage of the entire network is taken as the reference voltage of all DGs. In the ideal steady state, (18) can be changed into (27) Assuming that , is fixed, then the formula of DGs output active power is (28) By substituting droop coefficient expression into the above equation, it can be obtained that the actual power is precisely distributed according to generation command of dispatch (29) 4.2 Tie-line power optimisation control To mitigate the fluctuation of demand side and generation side, this paper proposes a control strategy of tie-line power by the feedback of power derivation. When a DG reaches the maximum, the corresponding DGA disengages coordinative consensus iteration automatically. Other DGs can undertake tie-line fluctuations in accordance with the original proportion determined by economy dispatch. The process is demonstrated in Fig. 2. Fig. 2Open in figure viewerPowerPoint Flow chart of tie-line power optimisation control As is mentioned in [31], the strategy that energy storage system adjusts all power deviations is called energy storage regulation. On the basis of CCA, a new adjustment mode of tie-line power is designed to reduce operating costs by seeking equilibrium according to the generation marginal costs. The process is as follows: Sample tie-line power and calculate power deviation ; Dealing with tie-line power deviation by low-pass filtering. Obtain the low-frequency component and send it to the DDGA; DDGA query MGCCA mode instructions. If storage adjustment mode is chosen, calculate the storage capacity. If storage capacity is greater than power deviation, only the energy storage adjusts deviation, and turn to step 4, otherwise turn to step 5; The MGCCA calculates the droop coefficient of each DG in real time by previous dispatch plan. The initial iteration value is set as the number of distributed generation units with available capacity and participates in the adjustment of tie-line power. Other initial states of those distributed generations are set as 0; After consensus iteration, all state values will converge to . Adjust DG reference power according to (30) to ensure that the total power is enough to stabilise line fluctuations, and the ratios of each DG power equals the economically optimised power proportion; A monitoring cycle ends after a delay. On the one hand, this method can respond to dispatch instructions. Thus, frequency regulation, peak shaving and other auxiliary services can be achieved by changing . On the other hand, the stochastic characteristics of power generation and load are reflected in the fluctuation of tie-line power. State observer feedbacks power deviation, changes the state value of DDGA and optimises power dispatch by consensus iteration. Thus, less capacity configuration of storage is required in this strategy. At the same time, ED reference can be traced accurately. The multi-agent collaborative optimisation, synergetic control based on CCA and droop control are divided into economy dispatch layer, secondary control layer and primary control layer. MG optimisation control model based on MAS can be summarised as shown in Fig. 1. 4.3 'Plug and play' distributed control MG is so widely distributed that the design of 'plug and play' distributed control method requires strong scalability. The ideal situation is that managers only need to modify a small amount of software parameters for a new access to the MG. Centralised control is demanding in real-time communication. In contrast, distributed consensus control uses line to communicate with consensus iteration, which is reliable and easy to extend. For the case of a new power generation node, the following steps can be used to extend net topology: (a) Recalculate and modify the number of adjacent nodes. (b) Neighbouring nodes send the numbers of neighbour nodes to each other. Find the maximum number to correct double random matrix. (c) Upgrade optimal scheduling function of MGCCA to achieve holistic dispatch according to new net topology. 5 Simulation and results To verify the optimisation control method proposed in this paper, a simulation model on pscad-4.5x platform is built. Agent functions are simulated by C programing in real time. The example topology is illustrated in Fig. 3. Fig. 3Open in figure viewerPowerPoint Case study of MG Set up five kinds of distributed generation, namely micro-gas turbine, photovoltaic, energy storage devices, fuel cells and wind power. The ADN's rated voltage is 10 kV while the rated voltage of MG is 0.4 kV. See Table 1 for transmission line parameters. Table 1. Transmission line parameters of case study Line no. Length, m Resistance, Ω Reactance, mH 1-2 50 0.0163 0.0116 2-3 300 0.0975 0.0696 3-4 200 0.0652 0.0464 4-5 100 0.0326 0.0232 2-6 250 0.0815 0.058 6-7 150 0.0498 0.0348 7-8 100 0.0326 0.0232 5.1 Results of economy dispatch The parameters of the optimisation model are illustrated in Table 2. While, forecasting load and time-of-use price are shown in Table 3. Other parameters are set as , , . Table 2. Parameters of the optimisation model Installation cost, $/kW Life, a Maintenance cost, $/kW/a Capacity DG1 3426 10 6.156 200 kWh DG2 1428 10 5.674 40 kW DG3 1242 20 0.923 30 kW DG4 2785 10 1.533 40 kW DG5 1714 15 3.247 65 kW Table 3. Electricity price and load forecasting in MG Period, h Price, ¥ Load, kW Period, h Price, ¥ Load, kW 1 0.17 28.3 13 0.83 101.9 2 0.17 37.2 14 0.83 104.3 3 0.17 39.2 15 0.83 100.4 4 0.17 38.6 16 0.49 91.1 5 0.17 50.2 17 0.49 84.9 6 0.17 58.5 18 0.49 88.0 7 0.17 64.7 19 0.83 101.1 8 0.49 52.0 20 0.83 106.7 9 0.49 68.3 21 0.83 111.4 10 0.49 75.2 22 0.49 96.2 11 0.83 84.3 23 0.49 57.8 12 0.83 92.4 24 0.17 51.2 According to MGCCA upper-level optimal dispatch model, the result of tie-line power dispatch by CGA and the internal dispatch of MG are shown in Figs. 4a and b, respectively. In Fig. 4a, the expected powers of MG and ADN represent the convergence points of the two population during the evolutionary process. The error between the two results can be limited by reducing . Fig. 4Open in figure viewerPowerPoint Upper-level economic scheduling results (a) Results of tie-line power game algorithm, (b) Economy dispatch results in MG The traditional MG optimisation aims at minimising the internal cost of MG, and the distributed network is used as the source to interact power with a fixed price. However, in the actual situation, ADN ancillary services, especially providing reserve capacity and peak shaving, should also be fully considered. The model proposed in this paper not only quantifies the cost of ancillary services, but also simulate the game process between MG and ADN. Compared with the traditional method, we can find that our scheduling strategy has a 2.36% advantage over the total cost (see in Table 4). Table 4. Economic comparison between two methods (1 day) Objective Model Cost of ADN, ¥ Cost of MG, ¥ Total cost, ¥ MG model in Section 3.2 38.8086 410.9654 449.7740 ADN + MG game model in Sections 3.1 and 3.2 20.7043 418.6996 439.4039 5.2 Secondary control effects Under the condition of grid connection, the optimised parameters of distributed power generations from 6 a.m. to 8 a.m. after the upper optimisation are brought into the system simulation to verify the effect and precision of the multi-agent optimisation instruction in a two-layer control. The rated active power outputs of all distributed generations in each of the three periods are adjusted for observation convenience, which are shown in Table 5. Table 5. Rated power generation of DGs Type No. 6:00 7:00 8:00 ES DDGA 15 25 −15 MT DGA1 40 45 65 PV DGA2 0 5 5 FC DGA3 15 15 40 WT DGA4 5 10 15 Active power outputs, voltage waveforms and iteration simulation of average voltage in MG from 6 a.m. to 8 a.m. are shown in Fig. 5. In Fig. 5a, compared with the power sharing without secondary synergetic control (dotted line), more precise active power tracking can be realised by consensus control (solid line) regardless of voltage mismatch shown in Fig. 5b. The priority target of the voltage consensus iteration is obtaining the average voltage of MG in real time to modify the droop control. In the 1.8–3 s period with large distributed voltage difference, voltage reference is adjusted automatically by adding compensation, which realises the adaptive generation of different circumstances. Fig. 5Open in figure viewerPowerPoint Results of active power control results (a) Active power comparison, (b) Voltage, (c) Average voltage consensus iteration For the control strategy of tie-line power [31], energy storage undertakes all tie-line power deviation in traditional strategy. Limited to storage capacity, the traditional method only applies to smaller fluctuations on the tie-line. Therefore, two disturbances in Load 3 is designed. Load 3 adds 20 + j12 kVA at 0.9 s, while resects 15 + j6 kVA at 1.8 s. The P, Q waveforms of tie-line power by traditional storage adjustment and economic adjustment are shown in Fig. 6. Fig. 6Open in figure viewerPowerPoint Comparison of tie-line fluctuation by different adjustment strategies Due to the consensus iteration, the sensitivity of economic adjustment is slightly slower than energy storage adjustment. However, the marginal cost of storage is high, leading to high operating expenses of MG. Generally speaking, the adjustment strategy by energy storage is fast, but the adjustment range is limited by energy storage; the strategy is not economy well. Regarding bigger fluctuations in the tie-line for a long time, simple energy storage regulation could not achieve the desired res