A real-time dispatching method for microgrids based on robust model predictive control

Abstract

This paper discloses a real-time scheduling method for microgrids based on robust model predictive control. This method combines model predictive control (MPC) with robust optimization (RO). The proposed MPC-based robust strategy formulates MPC optimization as a robustness problem, enabling real-time energy scheduling in the microgrid while ensuring reliable operation. Furthermore, the MPC feedback update mechanism helps reduce the conservatism of RO. Furthermore, the proposed solution considers demand response (DR) of power loads to further reduce operating costs. The effectiveness of the proposed method is demonstrated through simulation experiments and comparisons with benchmarks.

Description

Technical Field

[0001] The present invention relates to the technical field of microgrids, and in particular to a real-time scheduling method for microgrids based on robust model predictive control. Background Art

[0002] With increasing energy depletion and environmental degradation, renewable distributed generation (RDG) has garnered significant attention. Microgrid technology has emerged to address the intermittency and uncertainty of RDG while promoting its penetration and regulation within power systems. Therefore, the development of microgrids is crucial for the green and low-carbon operation of future energy systems. Furthermore, economic energy management in microgrids is a relevant topic, particularly regarding real-time energy scheduling under the uncertainties inherent in RDG and source-load coordination within microgrids. MPC technology can enable real-time decision-making in microgrid operations.

[0003] While MPC can roll forward prediction information to improve optimal control performance, unavoidable prediction errors still create uncertainties. These uncertainties pose risks to the reliable operation of microgrids. Robust optimization (RO) can effectively address decision-making under uncertainty. RO ensures the feasibility of decisions and the stability of grid operations in any scenario within a predefined uncertainty set, while effectively reducing the conservatism of its solutions. Summary of the Invention

[0004] In view of the deficiencies of the above-mentioned background technology, the present invention proposes a real-time scheduling method for microgrids based on robust model predictive control.

[0005] The present invention comprises the following steps:

[0006] Step 1: Establish the basic structure model of the microgrid

[0007] The basic structure model of the microgrid includes photovoltaic power generation, wind power generation, diesel generator sets and a battery-based energy storage system. The microgrid is connected to the main power grid through a common coupling point and has an energy management system.

[0008] Step 2: Establish a real-time scheduling model based on MPC

[0009] The objective function is established with the lowest operating cost of the microgrid, which includes the cost of purchasing electricity from the grid, the power generation cost of the diesel generator, and the attenuation cost of the power storage system;

[0010] Establish system constraints, including facility capacity constraints, load demand response constraints, and power balance constraints:

[0011] Step 3: Establish a robust real-time scheduling strategy

[0012] The real-time energy scheduling model is expressed as:

[0013]

[0014] Where x represents the decision variable and u is the uncertainty variable;

[0015] The objective function aims to find the best decision that is robust to any scenario in the uncertainty set, thus ensuring the reliability of energy supply even in the worst case;

[0016] The selection of uncertainty set includes the following sub-steps:

[0017] Construct the wind-solar historical output pattern set Ω based on historical data;

[0018] Divide the collected historical data of wind and solar power output by day, and the expression of Ω is:

[0019]

[0020] Among them, P TD,WT T D Wind power output during the period, P TD,PV T D Photovoltaic output during the period, N s The number of days of historical data collected is;

[0021] Step 3.2:

[0022] Assume that the forecast output pattern set composed of the day-ahead forecast values of uncertain variables obtained by the forecast model is:

[0023]

[0024] Step 3.3:

[0025] Scenarios with similar distribution to the predicted output pattern are selected from the historical output pattern to form the uncertainty set corresponding to the robust optimization period.

[0026] Step 3.4

[0027] Select uncertainty scenarios that meet the risk margin.

[0028] Beneficial Effects of the Invention: In this invention, MPC optimization is formulated as a robust problem, combining the advantages of both MPC and RO. This allows for real-time scheduling of microgrids in a rolling manner to meet current operational constraints. Furthermore, robust optimization is solved by filtering uncertainty sets based on KL divergence to ensure reliable microgrid operation. Furthermore, MPC's real-time feedback update mechanism effectively reduces the conservatism of RO solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a schematic diagram of the microgrid structure;

[0030] Figure 2 This is the basic principle diagram of MPC technology;

[0031] Figure 3 Predicted power distribution and base load diagram for PV and WT;

[0032] Figure 4 This is an overview of electricity prices;

[0033] Figure 5 The load power comparison diagram before and after demand response;

[0034] Figure 6 FIG4 is a comparison diagram of transmission power between the present invention and the benchmark. DETAILED DESCRIPTION

[0035] The present invention is further described below with reference to the accompanying drawings and embodiments:

[0036] Step 1: Establish the basic structure model of the microgrid

[0037] like Figure 1 As shown, the energy management system in this embodiment integrates multiple energy sources into an AC microgrid, including photovoltaic (PV), wind turbines (WT), diesel generator sets (TUs), and a battery-based energy storage system (ESS). The solid lines in the figure represent power flow between the source, grid, load, and storage, while the dashed lines represent information flow.

[0038] The microgrid is connected to the main grid through a common coupling point (PCC) to ensure the reliability of power supply. More specifically, the generator set includes non-dispatchable RDG (e.g., photovoltaic power generation PV and wind power generation WT) and diesel generators (TUs). A battery-based energy storage system (ESS) can store electricity during low-price periods for subsequent use, thereby reducing the operating costs of the microgrid. The power load can be converted into a signal that changes with the electricity price through demand response (DR) to further reduce operating costs.

[0039] Step 2: Establish a real-time scheduling model based on MPC

[0040] The basic principle of MPC is as follows Figure 2As shown in Figure 1, the optimization problem is executed at time t based on forecast information about uncertain variables over a period of time in the future, but only the decision solution at time t is executed. At time t+1, the previous process is repeated using the updated forecast information until the end of the optimization range. The advantage of this rolling optimization mechanism is that it can obtain more accurate forecast information every hour based on updated historical observation data, effectively improving the accuracy of optimal control.

[0041] According to the MPC principle, the real-time economic dispatch model of the microgrid based on prediction information can be formulated as follows:

[0042] (1) Objective function

[0043] The goal of the microgrid real-time dispatch model is to minimize the operating cost of the microgrid by coordinating the control of the microgrid's internal facilities.

[0044] The optimization objective function is:

[0045] min F=C grid +C diesel +C ESS (1)

[0046] Taking the lowest operating cost of microgrid as the objective function, C grid 、C diesel and C ESS They represent the cost of purchasing electricity from the grid, the power generation cost of the diesel generator, and the attenuation cost of the ESS, respectively.

[0047] In some embodiments, the cost of electricity purchased from the grid is calculated as follows:

[0048]

[0049] Where t represents a moment; t st Indicates the start time of the current MPC retreat, T indicates the total duration of the optimization range; λ t and P t,grid are the electricity price and transmission power between the grid and the microgrid at time t, respectively. It is assumed that the microgrid buys and sells electricity to the grid at the same price.

[0050] In some embodiments, the cost of electricity generation from a diesel generator is expressed as a quadratic function of its output power.

[0051]

[0052] Among them, P t,diesel represents the output power of the diesel generator at time t; a, b, c are the cost coefficients of the generator.

[0053] In some embodiments, the attenuation cost of the ESS is calculated as follows:

[0054]

[0055] Where ρ is the decay cost coefficient of ESS; P t,ch and P t,dis They represent the charging and discharging power of ESS at time t respectively.

[0056] (2) System constraints

[0057] In order to ensure the reliable operation of the microgrid, constraints must be met, including facility capacity constraints, load DR constraints, and power balance constraints.

[0058] In some embodiments, the power transmitted between the microgrid and the main grid needs to be limited to a certain range:

[0059]

[0060] in is the maximum value of the transmission power.

[0061] In some embodiments, the output power limit of the diesel generator and its ramp constraint are given in and , respectively:

[0062]

[0063]

[0064] in Indicates the maximum output power of the generator; R dn and R up Indicates the descending or ascending slope respectively.

[0065] In some embodiments, the dynamic model and operating constraints of the ESS are expressed as follows:

[0066]

[0067]

[0068]

[0069] SoC min ≤SoC t ≤SoC max (11)

[0070] in is the maximum charge or discharge power of ESS; SoC t represents the state of charge (SOC) of the ESS at time t and is constrained to protect the ESS; η ch and η disare the charging and discharging efficiencies, respectively; E represents the rated capacity of the ESS. It should be noted during implementation that binary variables or charge-discharge complementarity constraints that prevent simultaneous charging and discharging can be avoided here, as the objective function includes minimizing the ESS degradation cost.

[0071] In some embodiments, the power balance constraint requires that the power supplied from all resources must equal the power demand at every moment, as shown below:

[0072] P t,grid +P t,diesel +P t,ch +P t,dis +P t,PV +P t,WT =P t,ch +P t,load (12)

[0073] Among them, P t,PV and P t,WT are the uncertain power generated by PV and WT at time t; P t,load Represents the electrical load power at time t.

[0074] In some embodiments, the load within the microgrid can change with changes in electricity prices, thereby further reducing operating costs. This is called DR (Demand Response). DR features are as follows:

[0075]

[0076]

[0077] in and are the lower and upper bounds of the load in DR at time t, respectively; this constraint indicates that the total power demand needs to be met within the entire optimization range.

[0078] Step 3: Establish a robust real-time scheduling strategy

[0079] To address the prediction errors generated by model predictive control (MPC), a real-time energy scheduling strategy for microgrids based on MPC was designed by combining robust optimization with MPC technology. This robust model predictive control (RMPC)-based solution can make decisions while ensuring that current operational constraints and future energy supply reliability are met.

[0080] According to the robust optimization theory, the compact form of the real-time energy scheduling model based on RMPC designed in this implementation is expressed as:

[0081]

[0082] Where x represents the decision variables, including P t,grid , P t,diesel , P t,ch , P t,dis and P t,load ; u is an uncertainty variable, including P t,PV and P t,WT .

[0083] The objective function aims to find the optimal decision that is robust to any scenario within a predefined uncertainty set, thereby ensuring energy supply reliability even in the worst-case scenario. Therefore, the definition of the uncertainty set is crucial. This embodiment characterizes the likely occurrence of uncertain variables in a given time period by matching historical similarity patterns.

[0084] In a preferred embodiment, the selection of the uncertainty set includes the following sub-steps:

[0085] Step 3.1:

[0086] Construct wind (P t,WT )-light(P t,PV ) Historical output mode set Ω. The collected wind-solar output historical data is divided by day, and the number of days of collected historical data is set as N s , the total number of time periods in a day is T D . Then the expression of Ω can be written as:

[0087]

[0088] Step 3.2:

[0089] Assume that the forecast output pattern set composed of the day-ahead forecast values of uncertain variables obtained by the forecast model is:

[0090]

[0091] in and is the predicted value of the prediction model at time t. There are many methods and models for wind and solar power prediction, such as those based on long short-term memory networks. The prediction model can be selected based on merit in practical applications and will not be elaborated here.

[0092] Step 3.3:

[0093] Scenarios with similar distributions to the predicted output patterns are selected from the historical output patterns to form the uncertainty set for the corresponding robust optimization period. The adopted scenario screening method is candidate sorting based on the KL divergence (Kullback–Leibler divergence), that is, the degree of match between each historical scenario and the target pattern is measured by measuring the KL divergence between the historical output pattern set Ω and the predicted output pattern set Ω*. The smaller the KL divergence, the closer the distance between the two distributions and the greater the similarity. The discrete form calculation formula is as follows:

[0094]

[0095] Step 3.4

[0096] Based on step 3.3, the historical scenarios and target scenarios are sorted by KL divergence to select the uncertainty scenarios that meet the risk margin. The uncertainty set of the uncertainty variables involved in robust optimization in model predictive control meets the following conditions:

[0097]

[0098] where u t and is the uncertain variable at time t and the corresponding predicted value; Δu t is u t The maximum prediction error of t is the risk margin of uncertainty at time t.

[0099] The simulation results are as follows:

[0100] The effectiveness of the present invention is demonstrated by comparing it with the traditional robust optimization day-ahead scheduling strategy in simulation. All models are coded in MATLAB and solved using the Gurobi 9.5 solver.

[0101] The structure of the microgrid used is as follows Figure 1 The rated parameters of the facilities are shown in the table below. The duration of the moment is set to 1 hour, Δt = 1, and the total scheduling period is set to 24 hours, so T = 24.

[0102]

[0103] Figure 3 The forecast power distribution of PV and WT for a typical day is shown. The PV and WT forecast errors are set to 10% of the forecast value for the 4-hour period around the current MPC rolling cycle, and 20% for the remaining hours. The uncertainty budgets for PV and WT are set to 4 and 6, respectively; the nominal value of the base load is also shown. Figure 3 The load power is allowed to be 0.8-1.2 of its nominal value.

[0104] In addition, the electricity prices for simulation are as follows Figure 4 The actual load distribution after implementing DR is shown in Figure 5 As shown in Figure 2, it can be seen that the power load has changed compared to the nominal value. The interactive power curves of the present invention and the benchmark are shown in Figure 2. Figure 6 It can be observed that the input power of the microgrid in the baseline is higher than that of the present invention at all times, where the negative power represents the output power of the microgrid to the grid.

Claims

1. A real-time scheduling method for microgrids based on robust model predictive control, characterized by The method comprises the following steps: Step 1: Establish the basic structure model of the microgrid The basic structure model of the microgrid includes photovoltaic power generation, wind power generation, diesel generator sets and a battery-based energy storage system. The microgrid is connected to the main power grid through a common coupling point and has an energy management system. Step 2: Establish a real-time scheduling model based on model predictive control (MPC) The objective function is established with the lowest operating cost of the microgrid, which includes the cost of purchasing electricity from the grid, the power generation cost of the diesel generator, and the attenuation cost of the power storage system; Establish system constraints, including facility capacity constraints, load demand response constraints, and power balance constraints: Step 3: Establish a robust real-time scheduling strategy The real-time energy scheduling model is expressed as: Where x represents the decision variable and u is the uncertainty variable; The objective function aims to find the best decision that is robust to any scenario in the uncertainty set, thus ensuring the reliability of energy supply even in the worst case; The selection of uncertainty set includes the following sub-steps: Construct the wind-solar historical output pattern set Ω based on historical data; Divide the collected historical data of wind and solar power output by day, and the expression of Ω is: in T D Wind power output during the period, T D Photovoltaic output during the period, N s The number of days of historical data collected is; Step 3.2: Assume that the forecast output pattern set composed of the day-ahead forecast values of uncertain variables obtained by the forecast model is: Step 3.3: Select scenarios with similar distribution to the predicted output pattern from the historical output pattern to form the uncertainty set corresponding to the robust optimization period; Step 3.4 Select uncertainty scenarios that meet the risk margin.

2. A microgrid real-time scheduling method based on robust model predictive control according to claim 1, characterized in that: The cost of purchasing electricity from the grid C grid The calculation is as follows: Where t represents the time; t st Indicates the start time of the current MPC backoff, and T indicates the total duration of the optimization range; λ t and P t,grid are the electricity price and transmission power between the grid and the microgrid at time t, respectively, and △t represents the duration between two adjacent moments.

3. The method for real-time scheduling of a microgrid based on robust model predictive control according to claim 1, characterized in that: The cost of electricity generated by a diesel generator is C diesel The calculation is as follows: Among them, P t,diesel represents the output power of the diesel generator at time t; a, b, c are the cost coefficients of the generator, t st Indicates the start time of the current MPC backoff, and T indicates the total duration of the optimization range; △t represents the duration between two adjacent moments.

4. The method for real-time scheduling of a microgrid based on robust model predictive control according to claim 1, characterized in that: The decay cost C of the power storage system ESS The calculation is as follows: where ρ is the decay cost coefficient of the power storage system, P t,ch and P t,dis denote the charging and discharging power of the power storage system, t st Indicates the start time of the current MPC backoff, and T indicates the total duration of the optimization range; △t represents the duration between two adjacent moments.

5. The method for real-time scheduling of a microgrid based on robust model predictive control according to claim 1, characterized in that: The capacity constraints of the facility include transmission power constraints between the microgrid and the main grid, output power constraints of the diesel generator, output power ramp constraints of the diesel generator, and energy storage system operation constraints.

6. A microgrid real-time scheduling method based on robust model predictive control according to any one of claims 2 to 4, characterized in that: The decision variables x include P t,grid , P t,diesel , P t,ch , P t,dis and P t,load ; Uncertain variables u include P t,PV and P t,WT .

7. A microgrid real-time scheduling method based on robust model predictive control according to claim 6, characterized in that: The mode scenario screening adopted is based on KL divergence candidate ranking.

8. The method for real-time scheduling of a microgrid based on robust model predictive control according to claim 7, characterized in that: The uncertainty set of the uncertainty variables involved in robust optimization in model predictive control satisfies the following conditions: where u t and is the uncertain variable at time t and the corresponding predicted value; △u t is u t The maximum prediction error of Γ t is the risk margin of uncertainty at time t, and the superscript RDG represents renewable distributed generation.

Images