A multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scenario prediction

Abstract

The application discloses a kind of real-time energy regulation methods of multi-energy microgrid based on dynamic tubular model predictive control and data-driven scene prediction belonging to the technical field of multi-energy microgrid energy regulation.It includes: based on the prediction of multi-energy microgrid operation scene by Wasserstein generative adversarial network;Establish nominal MPC model, obtain the reference trajectory of multi-energy microgrid operation scene prediction;Based on actual data and reference trajectory, determine the tubular region on short time scale;Establish auxiliary MPC model, track the reference trajectory in tubular region, formulate multi-energy microgrid actual operation scheme, until the next long time scale time, according to the latest system operating state re-performs scene prediction, and is passed to nominal MPC model;Nominal MPC model dynamically corrects reference trajectory, re-determines tubular region, and uses auxiliary MPC model to carry out real-time energy regulation.The application guarantees to meet the operation constraint of multi-energy microgrid and improves its operation economy under the interference of uncertainty.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multi-energy microgrid energy regulation, and particularly relates to a multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scene prediction. BACKGROUND

[0002] In recent years, with the intensification of fossil energy crisis and environmental deterioration, photovoltaic and wind power and other renewable distributed generation have been greatly promoted. In this regard, multi-energy microgrid (MEMG) can well integrate renewable distributed generation with power systems, and it can utilize the flexibility of internal multiple energy forms to realize multi-energy complementation, thereby improving energy utilization efficiency and promoting the consumption of renewable distributed generation. Therefore, the optimal energy management of multi-energy microgrid has attracted great attention, which is the key to realizing the environmentally friendly, energy-saving and economic operation of multi-energy microgrid. However, the intermittency and randomness of renewable distributed generation and multi-energy load will cause the uncertainty of multi-energy microgrid operation, which will pose great challenges to the reliability and economy of multi-energy microgrid energy management strategy.

[0003] For the uncertainty of multi-energy microgrid operation, the existing multi-energy microgrid energy management strategy adopts technologies including robust optimization, stochastic programming and chance constraint, etc. However, the aforementioned technologies make decisions based on day-ahead information, and their performance will be greatly limited by the inevitable difference between the day-ahead information and the actual dynamic uncertainty, which may lead to suboptimal energy management decisions. Although model predictive control (MPC) can realize real-time energy regulation, this technology still highly depends on accurate prediction of uncertainty. Other existing technologies (such as robust model predictive control, stochastic model predictive control) will still lead to overly conservative multi-energy microgrid energy management decisions or failure to meet operational constraints, etc., and stochastic model predictive control needs to generate a set of operational scenarios at each time that can accurately depict operational uncertainty. However, existing scene prediction technologies mostly use random sampling from pre-assumed statistical distribution to obtain operational scenarios, but the assumed statistical distribution often cannot be maintained in practice, thereby reducing the representativeness of predicted scenarios and damaging the performance of stochastic model predictive control.

[0004] Therefore, there is a need for a multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scene prediction, which can guarantee to meet the operational constraints of multi-energy microgrid and improve its operational economy under the interference of uncertainty. SUMMARY

[0005] The application aims to provide a multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scene prediction, comprising the following steps:

[0006] The multi-energy microgrid operation scene is predicted based on a Wasserstein generative adversarial network;

[0007] A nominal MPC model is established to obtain a reference trajectory for multi-energy microgrid operation scene prediction;

[0008] Based on actual data and the reference trajectory, a tubular region is determined on a short time scale;

[0009] An auxiliary MPC model is established to track the reference trajectory in the tubular region, formulate a multi-energy microgrid actual operation scheme, until the next long time scale time, re-perform scene prediction according to the latest system operation state, and pass to the nominal MPC model;

[0010] The nominal MPC model dynamically corrects the reference trajectory, re-determines the tubular region, and uses the auxiliary MPC model to perform real-time energy regulation.

[0011] Further, the nominal MPC model comprises:

[0012] Objective function:

[0013]

[0014] Wherein and respectively represent the expected value of the power and natural gas procurement cost and the ESS and HSS degradation cost under all predicted operation scenes at time k; π s is the occurrence probability of each scene; and are the purchase and sale prices of electricity to the power grid; and respectively represent the power purchase and sale power at time t under scene s; Δt is the time interval of the next time under the long time scale; λ gas is the natural gas procurement price; represents the input gas power of CHP; ρ ESS and ρ HSS are the degradation cost coefficients of ESS and HSS; and respectively represent the discharging and charging power of ESS; and represent the heat release and heat charging power of HSS; N s is the number of predicted scenes; T is the optimization time range;

[0015] Constraint condition:

[0016] Power balance constraint

[0017]

[0018] where, is the interaction power between MEMG and main grid at time t in scenario s; and are the electrical power of PV generation, WT generation and electrical load, respectively; and are the electrical power of CHP output and EB consumption, respectively; and are the thermal power of CHP and EB output, respectively; is the thermal load power;

[0019] Transmission power constraint

[0020]

[0021] where, is the upper limit of transmission power between MEMG and main grid; is a binary variable for avoiding simultaneous purchase and sale of electricity;

[0022] Energy conversion equipment operation constraint

[0023]

[0024] where, is the upper limit of η CHP is the comprehensive conversion efficiency of CHP; κ and are the lower and upper limits of CHP thermal-to-electricity ratio, respectively; ΔP gCHP is the ramping power limit of CHP; η EB is the electrical-to-thermal efficiency of EB;

[0025] Energy storage equipment operation constraint

[0026]

[0027]

[0028] where, is the stored energy in ESS or HSS at time t in scenario s; and are the charging or discharging efficiency; is a binary variable for avoiding simultaneous charging and discharging; Q i and are the lower and upper limits of stored energy in ESS or HSS, respectively; ΔQ ito store the extent of tightening of the energy storage constraint; denotes the operating power of the ESS or HSS.

[0029] Further, the auxiliary MPC model comprises:

[0030] Objective function:

[0031]

[0032] wherein, and respectively denote the power and natural gas procurement cost due to decision adjustment at time instance l, and the degradation cost of ESS and HSS; ΔP l buy and ΔP l sell respectively denote the power purchase and sale power adjustment of MEMG; Δt s is the short time scale time interval; ΔP l gCHP denotes the input power adjustment amount of CHP; ΔP l ESS,d and ΔP l ESS,c respectively denote the discharge and charge power adjustment amount of ESS; ΔP l HSS,d and ΔP l HSS,c respectively denote the heat release and heat charging power adjustment amount of HSS; n is the prediction time range at each adjustment decision-making time; t s is the starting time of the auxiliary MPC model regulation;

[0033] Tubular region range determination sub-model:

[0034]

[0035] wherein, ΔP l and Δh l respectively denote the electric-thermal imbalance power at time instance l; and respectively denote the electric power reference operating decision of ESS, CHP and EB at time instance k; and respectively denote the thermal power reference operating decision of CHP, EB and HSS; P l PV , P l WT , P l load and respectively denote the ultra-short-term predicted power of PV power generation, WT power generation, electric load and thermal load at time instance l; Ω kto assist the MPC to correspond to the nominal MPC of the time set of k;

[0036] Constraints:

[0037] Power balance constraints

[0038] -ΔP l = ΔP l grid + ΔP l ESS,d + ΔP l eCHP - (ΔP l ESS,c + ΔP l eEB

[0039] - Δh l = ΔP l hCHP + ΔP l hEB + ΔP l HSS,d - ΔP l HSS,c

[0040] wherein, ΔP l grid is the power adjustment amount of the time lMEMG and the grid interaction; ΔP l eCHP and ΔP l eEB respectively represent the CHP output electric power and EB input electric power adjustment amount; ΔP l hCHP and ΔP l hEB respectively represent the CHP and EB output heat power adjustment amount;

[0041] Transmission power adjustment constraints

[0042]

[0043] ΔP l grid = ΔP l buy - ΔP l sell

[0044]

[0045] wherein, is a binary variable for avoiding power purchase and sale contemporaneous adjustment;

[0046] Energy conversion equipment operation adjustment constraints​

[0047] -|ΔP l |≤ΔP l i ≤|ΔP l |,i∈{eCHP,hCHP,hEB}

[0048]

[0049] ΔP l eCHP +ΔP l hCHP =η CHP ΔP l gCHP

[0050]

[0051] Energy storage device operation adjustment constraint

[0052]

[0053]

[0054] wherein, is the reference storage energy of the ESS and HSS at time k, and is a binary variable for avoiding simultaneous adjustment of the charge and discharge decisions of the ESS and HSS; and respectively represent the energy stored by the energy storage device i at time l and l-1; ΔP l i,c and ΔP l i,d respectively represent the charge and discharge power adjustment amount of the energy storage device i at time l.

[0055] The beneficial effects of the present application are that:

[0056] 1. The present application, aiming at the operation uncertainty of a multi-energy microgrid, based on two cascaded nominal MPC and auxiliary MPC models under a tubular model predictive control framework, deals with the operation uncertainty of the MEMG at different time scales, and realizes real-time formulation of the MEMG operation scheme under the premise of meeting all operation constraints.

[0057] 2. Compared with the existing technologies for multi-energy microgrid energy management based on day-ahead information, such as robust optimization, stochastic programming and chance constraint, the present application avoids the performance degradation caused by the inevitable difference between the previous day information and the actual uncertainty dynamics.

[0058] 3、The application describes the operation uncertainty by a group of operation scenes, avoids the high dependence on point prediction accuracy of traditional MPC technology and the overly conservative operation scheme decision of robust MPC, and improves the economy and accuracy of real-time decision of MEMG operation.

[0059] 4、The data-driven MEMG operation scene prediction method based on WGAN in the application avoids the drawbacks of existing scene prediction technology, i.e., random sampling from a pre-assumed statistical distribution to obtain an operation scene, the assumed statistical distribution cannot be maintained in practice, thereby reducing the representativeness of the predicted scene and damaging the performance of random MPC. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 The figure is a flowchart of the multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scene prediction.

[0061] Figure 2 The figure is a schematic diagram of the overall framework of the multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scene prediction.

[0062] Figure 3 It is the electric-thermal unbalanced power at each time.

[0063] Figure 4 (a) (b) (c) are respectively the nominal and actual control trajectories of ESS, HSS and interactive power.

[0064] Figure 5 (a) (b) are respectively the SOC of ESS and the change diagram of HSS energy storage thermal energy.

[0065] Figure 6 (a) (b) are respectively the actual operation plan diagram of electric power and thermal power of MEMG. DETAILED DESCRIPTION

[0066] The application provides a multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scene prediction.

[0067] Figure 1 The figure is a flowchart of the multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scene prediction; Figure 2The schematic diagram of the overall framework of the multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scenario prediction of the application is shown in the figure. The nominal MPC generates a reference curve in a time sequence backward manner according to the characterization of uncertainty and tightening constraints in a long time scale. The developed nominal MPC incorporates the idea of stochastic programming, which characterizes the uncertainty of the multi-energy microgrid at each time through a set of multi-energy microgrid operation scenarios.

[0068] Step 1: predicting the multi-energy microgrid operation scenario based on the Wasserstein generative adversarial network;

[0069] The sources of MEMG operation uncertainty include renewable distributed power generation and multi-energy load. Taking photovoltaics (PV) and wind turbines (WT) power generation and electric heating load as examples, a MEMG operation scenario that can capture the MEMG operation uncertainty at future k~T time can be expressed as formula (1):

[0070]

[0071] Among them and represent the electric power of PV and wind turbine power generation and electric load at k time, respectively; is the power of the heating load; T is the time range of MEMG operation.

[0072] When predicting the uncertainty of MEMG at future k~T time, the available information about uncertain variables includes the observed values at past k-1 time and the point prediction values at future k~T time. Therefore, the scenario prediction method includes two parts as shown in Figure 1 : first, based on the historical operation scenario of MEMG, a WGAN is trained to obtain an unsupervised representation of its statistical distribution, and then based on the trained WGAN, an optimization problem is established at each time using available information to optimize the input noise vector. The training process of WGAN and the scenario prediction optimization problem are described as follows:

[0073] WGAN consists of a generator G and a discriminator D, and they use the Wasserstein distance as the value function of the game between the two. Assuming that x is a historical scenario and follows an unknown statistical distribution P x . Let z be a random input hidden vector sampled from a simple known distribution, and the output scenario obtained by converting it through the generator is represented as G(z), and its statistical distribution is P Gis performed. The discriminator D takes x or G(z) as input and its output D(x) or D(G(z)) is an indication of whether the scene is true or false. Thus, the game between G and D based on Wasserstein distance can be represented as (2). G is trained to convert z into a MEMG running scene as realistic as possible so as to make D believe that the generated scene is from a real historical scene as much as possible. D is trained to distinguish P x and P G , so as to distinguish the generated and historical MEMG running scenes. According to the training objectives of G and D, their training loss functions can be defined as (3) and (4) respectively.

[0074]

[0075] where E denotes the empirical mean in the training process.

[0076] When G and D are alternately trained to the Nash equilibrium of the game between them, the WGAN is considered to be trained, so as to establish an unsupervised representation of the statistical distribution of the MEMG running scene through the output of G. Thus, G can convert the random hidden vector sampled from P z into a MEMG running scene that looks real. In order to follow the available information to predict the MEMG running scene at k~T time, the present application establishes the following optimization problem to optimize the input vector of G:

[0077]

[0078] s.t.z∈Z(6)

[0079]

[0080] where P obse (G(z)) and P pred (G(z)) represent the part of G(z) at k-1 past time and the remaining k~T time; γ is a weight factor. The objective function (5) includes two terms, wherein the first term represents that the part of the predicted scene at k-1 past time should be as close as possible to the observed value s obse ; and the second term is a penalty term for ensuring that the predicted scene can maintain the statistical characteristics of the historical scene. The constraint (6) defines the definition domain Z of z. The constraint (7) limits the part of the predicted scene at the remaining k~T time not to fall outside the prediction interval constructed according to the point prediction value , so as to ensure that the predicted scene does not conflict with the point prediction information. The upper and lower bounds of the prediction interval and can be calculated according to (8) and (9) respectively.

[0081]

[0082]

[0083] where a is a predefined prediction confidence level and a > 1.

[0084] The objective function (5) includes G and D which are trained and highly non-convex, thus solving optimization problems (5)-(7) iteratively with different initial values z can lead to different local optimal solutions. Assume that optimization problems (5)-(7) are solved iteratively with N s different initial values z, N s local optimal solutions By inputting these local optimal solutions into the trained G, a set of N s apparently realistic future MEMG operation scenarios at k ~ T time instants can be predicted as shown in (10).

[0085]

[0086] Step 2: Establish a nominal MPC model to obtain the reference trajectory of the multi-energy microgrid operation scenario prediction;

[0087] To consider all potential MEMG operation scenarios, the nominal MPC model at time instant k can be established as (11)-(15), whose objective is to minimize the expected value of the cost under all operation scenarios in the prediction horizon. The operation cost includes the electricity purchase cost from the main grid, the natural gas procurement cost, and the degradation cost of the electricity storage system (ESS) and the heat storage system (HSS).

[0088]

[0089] where and represent the expected value of the electricity and natural gas procurement cost and the ESS and HSS degradation cost under all operation scenarios predicted at time instant k; p s is the occurrence probability of each scenario; and are the electricity purchase and sale prices from the grid; and represent the electricity purchase and sale power at time instant t under scenario s; At is the time interval of the next time instant under the long time scale; p gas is the natural gas procurement price; represents the input gas power of the combined heat and power unit (CHP); p ESS and pHSS The decay cost coefficients for ESS and HSS are respectively. and These represent the discharge and charging power of the ESS, respectively. and This indicates the heat release and heat charge power of the HSS.

[0090] The nominal MPC model must satisfy all operational constraints, including power balance constraints, transmission power limit constraints, and power conversion and storage device operation constraints, as described below:

[0091] (1) Power balance constraint: For electrical energy and thermal energy, the supply and demand should be balanced, as shown in equations (16) and (17).

[0092]

[0093] in Let be the interaction power between the MEMG and the main power grid at time t under scenario s; and These represent the electrical power generated by PV power generation, WT power generation, and electrical load, respectively. and These represent the output of the CHP and the electrical power consumed by the electric boiler (EB), respectively. and These represent the output thermal power of CHP and EB, respectively. This represents the heat load power.

[0094] (2) Transmission power limitation: To ensure the reliable operation of the main power grid, the power of the tie line needs to be limited to a certain range, and the purchase and sale of electricity by the MEMG cannot occur simultaneously, as shown in equations (18) to (19). In addition, the interactive power It can be expressed as equation (20).

[0095]

[0096] in This represents the upper limit of the power transmitted between the MEMG and the main power grid; This is a binary variable used to avoid simultaneous electricity purchase and sale.

[0097] (3) Operational constraints of energy conversion equipment: The operation of energy conversion equipment must meet the output power limit and energy coupling constraint. Equation (21) limits the output power range of each equipment. Equation (22) represents the relationship between the output electrical power and thermal power of CHP and its input power. Equation (23) represents that the thermoelectric ratio of CHP can be adjusted within a certain range. Equation (24) represents the ramp-up constraint of CHP. The thermal efficiency of EB is expressed in Equation (25).

[0098]

[0099] wherein denotes the upper limit of ; η CHP is the overall conversion efficiency of CHP; Kappa and are the lower and upper limits of CHP heat-to-power ratio, respectively; ΔP gCHP is the ramping power limit of CHP; η EB is the electrical-to-thermal efficiency of EB.

[0100] (4) Energy storage device operation constraints: The dynamic model of storage device and its operation constraints are shown in equations (26)-(30), where the relationship between the storage energy at two adjacent time instants is shown in equation (26), the charging and discharging power is subject to equations (27)-(28), and constraint (29) is used to avoid over-discharging and over-charging of storage device. In particular, the energy storage constraints in the nominal MPC model are tightened to leave some flexibility for the auxiliary MPC model to ensure the satisfaction of actual operation constraints. The operation power of ESS and HSS is calculated according to equation (30).

[0101]

[0102] wherein denotes the stored energy in ESS or HSS at time t under scenario s; and are the charging or discharging efficiency; is a binary variable used to avoid simultaneous charging and discharging; Q i and are the lower and upper limits of ESS or HSS stored energy, respectively; ΔQ i is the tightening degree of stored energy constraints; denotes the operation power of ESS or HSS.

[0103] By solving the nominal MPC model with objective functions (11)-(15) and constraints (16)-(30), the operation decisions at time k for all scenarios can be obtained. Then, the reference trajectory can be obtained by calculating the expected value of operation decisions under all scenarios. For example, , which can be calculated by equation (31).

[0104]

[0105] Step 3: Determine the tubular region on a short time scale based on actual data and reference trajectory;

[0106] Step 4: Establishing the auxiliary MPC model to track the reference trajectory in the tubular region, and to formulate the actual operation scheme of the multi-energy microgrid until the next long-time-scale time, according to the latest operation state of the system to re-perform scene prediction and pass to the nominal MPC model;

[0107] The purpose of the auxiliary MPC in the short-time scale is to track the reference trajectory formulated by the nominal MPC and to formulate the actual operation decision of the MEMG that meets all the operation constraints most economically under uncertain disturbances. Therefore, the objective function of the auxiliary MPC model can be formulated as equations (32) to (36).

[0108]

[0109]

[0110] wherein and are the power and natural gas procurement costs and the degradation costs of the ESS and HSS caused by decision adjustment at time l (short-time scale); ΔP l buy and ΔP l sell are the power purchase and sale power adjustments of the MEMG; Δt s is the time interval of the short-time scale; ΔP l gCHP represents the input power adjustment amount of the CHP; ΔP l ESS,d and ΔP l ESS,c respectively represent the discharge and charge power adjustment amounts of the ESS; ΔP l HSS,d and ΔP l HSS,c respectively represent the heat release and heat charging power adjustment amounts of the HSS; n is the prediction time range at each adjustment decision.

[0111] The decision adjustment of the auxiliary MPC is to be made in the tubular region according to the reference decision formulated by the nominal MPC and the actual dynamics of uncertainties. The tubular region refers to the range allowed by the decision variable adjustment, so that the decision is adjusted in the tube to ensure that the actual operation decision not only meets the operation constraints of the system, but also guarantees the global optimality of the decision. Therefore, the unbalanced power needs to be calculated first to determine the range of the tubular region, as shown in equations (37) to (38).

[0112]

[0113] wherein ΔP l and Δh l are the electric-thermal unbalanced power at time l; and are the reference operational decisions of ESS, CHP and EB at time k (long time scale) for electrical power; and are the reference operational decisions of CHP, EB and HSS for thermal power; P l PV , P l WT , P l load and are the ultra-short term predicted powers of PV generation, WT generation, electrical load and thermal load at time / ; Ω k is the set of time instants of the auxiliary MPC corresponding to the nominal MPC at time k.

[0114] The auxiliary MPC model must also satisfy all the operational constraints of the MEMG, as described below:

[0115] (1) Power balance constraints: all the decision adjustments at each time instant must correspond to unbalanced powers, as shown in equations (39)-(40).

[0116] - ΔP l = ΔP l grid + ΔP l ESS,d + ΔP l eCHP - ΔP l ESS,c + ΔP l eEB ) (39)

[0117] - Δh l = ΔP l hCHP + ΔP l hEB + ΔP l HSS,d - ΔP l HSS,c (40)

[0118] where ΔP l grid is the MEMG interaction power adjustment with the grid at time / ; ΔP l eCHP and ΔP l eEB represent the CHP electrical power output and EB electrical power input adjustment, respectively; ΔP l hCHP and ΔP l hEB represent the CHP and EB thermal power output adjustment, respectively.

[0119] (2) Transmission power adjustment constraints: The power purchase and sale adjustment cannot happen simultaneously and cannot exceed the power imbalance, as shown in Equations (41)-(42). In addition, ΔP l grid can be calculated according to Equation (43) and the adjusted interaction power still cannot exceed the transmission power limit of the tie-line, as shown in Equation (44).

[0120]

[0121] ΔP l grid = ΔP l buy - ΔP l sell (43)

[0122]

[0123] where is a binary variable for avoiding the simultaneous adjustment of power purchase and sale.

[0124] (3) Energy conversion equipment operation adjustment constraints: The energy conversion equipment operation decision adjustment amount also cannot exceed the imbalance power, as shown in Equation (45). Equation (46) indicates that the adjusted equipment operation power still cannot exceed the allowed operation range. Equations (47)-(49) and (50) constrain the CHP and EB to follow the energy coupling constraints during the decision adjustment process.

[0125] - | ΔP l | ≤ ΔP l i ≤ | ΔP l |, i e {eCHP, hCHP, hEB} (45)

[0126]

[0127] ΔP l eCHP + ΔP l hCHP = η CHP ΔP l gCHP (47)

[0128]

[0129] (4) Energy storage equipment operation adjustment constraints: In the auxiliary MPC model, the reference storage energy of the ESS and HSS at time k The actual energy storage change is calculated according to equation (52). Equations (53)-(54) and (55)-(56) constrain the charging and discharging power adjustment range of ESS and HSS, respectively. Equations (57) and (58) indicate that the actual operating power and the energy that can be stored of ESS and HSS cannot exceed their original operating constraints, respectively.

[0130]

[0131]

[0132] wherein and are binary variables for avoiding simultaneous adjustment of charging and discharging decisions of ESS and HSS.

[0133] The operating adjustment decision of MEMG at time l can be obtained by solving the auxiliary MPC model shown in equations (32)-(58). The actual operating decision can be obtained by superimposing the adjustment decision on the reference decision made by the nominal MPC model, as shown in equation (59) (taking the transmission power between MEMG and the main grid as an example).

[0134]

[0135] Step 5: The nominal MPC model dynamically corrects the reference trajectory, re-determines the tubular region, and uses the auxiliary MPC model for real-time energy regulation.

[0136] In this embodiment, if the reference trajectory obtained by the nominal MPC is directly used for the operation of MEMG, due to the interference of operating uncertainty, it will lead to power imbalance, as shown in Figure 3 the thermal imbalance power of each period. These data will also be used to construct the deviation tube of the auxiliary MPC. The auxiliary MPC will adjust the reference trajectory in the tube in the most economical way to determine the actual control trajectory that meets all operating constraints at the same time. Figure 4 (a) (b) (c) are the nominal and actual control trajectories of ESS, HSS and interactive power, respectively. The positive and negative of the interactive power represent the purchase or sale of electricity, respectively. The positive and negative power of ESS and HSS represent that they are in the discharging or charging state, respectively. The changes of SOC of ESS and the stored thermal energy of HSS corresponding to this process are shown in Figure 5 (a) (b). Figure 4 From (a) (b) (c), it can be seen that by using the auxiliary MPC to cope with uncertainty, a more reasonable MEMG actual operating scheme is obtained. For example, negative power imbalance power (as shown in Figure 3 ) occurs in the first 5 periods, which means that MEMG is short of electricity. Therefore, ESS reduces its charging power (as shown in Figure 4(a) is shown). At this time, the HSS also switches to the exothermic state to compensate for the lack of MEMG heat power (as shown in Figure 4 (b). Figure 4 (c) shows that the interaction power is also adjusted according to the unbalanced power (for example, time #81-#85). Figure 5 The results in (a) (b) show that due to the tightening of the constraints, the nominal MPC will retain a certain dynamic margin. This allows the auxiliary MPC to exceed the tightened constraints during the adjustment process, but at the same time ensures that the original constraints are met to ensure the reliability of the scheduling plan (for example, time #105).

[0137] From Figure 4 (a) (b) (c) and Figure 5 It can also be seen from (a) (b) that the actual operation plan is obtained by tracking the reference trajectory in the deviation tube by the auxiliary MPC, which shows that the auxiliary MPC can still guarantee global optimality despite the shorter optimization time range. Figure 6 (a) (b) are respectively the actual operation plan diagrams of the electric power and heat power of the MEMG, from Figure 6 The global optimality of the decision result can also be seen from the actual MEMG operation scheme shown in (a) (b). Specifically, during the period when the electricity price is low (for example, time #1-#6 and #19-#21), the MEMG increases the purchase of electricity and stores the excess electricity in the ESS for subsequent use (as shown in Figure 6 (a)). In contrast, during the period when the electricity price is high, in addition to the RDG, the energy demand can be more economically met by consuming natural gas and releasing the stored electricity in the ESS (as shown in Figure 6 (a)). In addition, the MEMG also sells excess electricity to the main grid for arbitrage (for example, time #95). In addition, due to the adjustable characteristics of the heat-to-power ratio of cogeneration, it can maintain a large amount of heat output of cogeneration during low power output to ensure heat demand (as shown in Figure 6 (b).

[0138] In summary, the present application avoids the high dependence of traditional MPC technology on point prediction accuracy and the overly conservative operation scheme decision of robust MPC, improves the economy and accuracy of real-time formulation of MEMG operation decision, and avoids the drawbacks of existing scene prediction technology of randomly sampling from pre-assumed statistical distribution to obtain operation scenes.

Claims

1. A multi-energy microgrid real-time energy regulation method based on dynamic tubular model predictive control and data-driven scenario prediction, characterized in that, The method comprises the following steps: The multi-energy microgrid operation scenario is predicted based on a Wasserstein generative adversarial network; The scenario prediction comprises two parts: firstly, a WGAN is trained based on a historical operation scenario of the MEMG to obtain an unsupervised representation of a statistical distribution of the WGAN, and then an optimization problem is established based on the trained WGAN at each time using available information to optimize an input noise vector; A nominal MPC model is established to obtain a reference trajectory of the multi-energy microgrid operation scenario prediction; By solving the nominal MPC model, the operation decisions of all scenarios at time are obtained, and the reference trajectory is obtained by calculating the expected value of the operation decisions of all scenarios; Based on actual data and the reference trajectory, a tubular region is determined on a short time scale; Decision adjustment of the auxiliary MPC is performed within a tubular region established according to a reference decision made by the nominal MPC and actual dynamics of uncertainty, and the tubular region refers to a range allowed for adjustment of a decision variable; An auxiliary MPC model is established to track the reference trajectory within the tubular region, to formulate an actual operation scheme of the multi-energy microgrid, until a next long time scale, scenario prediction is performed again according to a latest operation state of the system, and is transmitted to the nominal MPC model; The nominal MPC model dynamically corrects the reference trajectory, re-determines the tubular region, and uses the auxiliary MPC model to perform real-time energy regulation. 2.The method of claim 1, wherein, The nominal MPC model comprises: An objective function: ， ， ， ， ， where , , and represent the expected values of the electricity and natural gas procurement costs and of the ESS and HSS degradation costs, respectively, over all the operating scenarios predicted at the instant; is the probability of occurrence of each scenario; and are the electricity purchase and sale prices to the grid, respectively; and represent the electricity purchase and sale powers at the instant in the scenario ; is the time interval of the next instant at the long time scale; is the natural gas procurement price; represents the input gas power of the CHP; and are the degradation cost coefficients of the ESS and HSS, respectively; and represent the discharging and charging powers of the ESS, respectively; and represent the discharging and charging powers of the HSS, respectively; is the number of predicted scenarios; is the optimization time horizon; Constraint conditions: Power balance constraint ， ，， wherein, is the scenario is the time is the interaction power of MEMG with the main grid at time t; , and are the electrical powers of PV generation, WT generation and electrical load, respectively; and are the electrical powers of CHP output and EB consumption, respectively; and are the thermal powers of CHP and EB output, respectively; is the thermal load power; Transmission power constraint ， ， ， wherein, is an upper limit for the transmission of power between the MEMG and the main grid; is a binary variable for avoiding simultaneous purchase and sale of electricity; Energy conversion device operation constraint ， ， ， ， ， wherein, represents an upper limit; is the overall conversion efficiency of the CHP; and are a lower limit and an upper limit, respectively, of the CHP thermal-to- electric ratio; is the ramping power limit of the CHP; is the electrical-to-thermal efficiency of the EB; Energy storage device operation constraint ， ， ， ， ， wherein, represents the scenario the ESS or HSS at the time instant t; and is the charging or discharging efficiency; is a binary variable for avoiding simultaneous charging and discharging; and are the lower and upper limits of the stored energy of the ESS or HSS, respectively; is the tightening degree of the stored energy constraint; represents the operating power of the ESS or HSS. 3.The method of claim 2, wherein, The auxiliary MPC model comprises: An objective function: ， ， ， ， ， in, , , and They are respectively The costs of electricity and natural gas procurement, as well as the degradation costs of ESS and HSS, arise constantly due to decision adjustments. and These refer to the power purchase and sales adjustments for MEMG; A time interval on a short time scale; This indicates the amount of input power adjustment for the CHP. and These represent the discharge and charging power adjustment amounts of the ESS, respectively; and These represent the adjustment amounts of heat release and heat charge power of the HSS, respectively. The forecast timeframe for each decision adjustment; The starting point for assisting in the regulation of the MPC model; A tubular region range determination submodel: ， ， wherein, and are the electrical and thermal unbalance powers at time ; , and are the electrical power reference operational decisions of the ESS, CHP and EB at time ; , and are the thermal power reference operational decisions of the CHP, EB and HSS; , , and are the ultra-short term predicted powers of the PV generation, WT generation, electrical load and thermal load at time ; is the set of time instants of the auxiliary MPC corresponding to the nominal MPC at time ; Constraint conditions: Power balance constraint ， ， wherein, is the time instant MEMG is the power adjustment amount of the interaction between the MEMG and the grid; and respectively represent the CHP output electric power and the EB input electric power adjustment amount; and respectively represent the CHP and EB output thermal power adjustment amount; Transmission power adjustment constraint ， ， ， ， wherein, is a binary variable for avoiding power purchase and sale colleague adjustment; Energy conversion device operation adjustment constraint ， ， ， ， ， ， Energy storage device operation adjustment constraint The nominal MPC model comprises: An objective function: Constraint conditions: Power balance constraint Transmission power constraint Energy conversion device operation constraint Energy storage device operation constraint The auxiliary MPC model comprises: An objective function: A tubular region range determination submodel: Constraint conditions: Power balance constraint Transmission power adjustment constraint Energy conversion device operation adjustment constraint Energy storage device operation adjustment constraint ， ， ， ， ， ， ， ， wherein, ESS and HSS at time of reference stored energy, and is a binary variable for avoiding simultaneous adjustment of charge-discharge decision of ESS and HSS; and respectively represent the stored energy of energy storage device at time and ; and respectively represent the charge-discharge power adjustment amount of energy storage device at time .

Images