A Distributed Real-Time Economic Dispatch Method and Device for a Microgrid System

Abstract

The present invention discloses a distributed real-time economic dispatch method and device for a microgrid system, belonging to the technical field of electrical engineering. The method includes: establishing an intraday real-time dispatch model of the microgrid system and reconstructing it into a distributed Markov decision process; using a piecewise linear function to approximately describe the approximate optimal value function from time t to the total dispatch domain T in state S t ; proposing a distributed value function training method based on a consensus-based distributed optimization algorithm to perform distributed training on the piecewise linear function; determining the state variable S t at time t based on the random factor R t of the microgrid system at time t, solving the Markov decision process based on the trained piecewise linear function to obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t, and further implementing the distributed dispatch of the microgrid system. The present invention can reduce communication and computational losses and ensure system privacy, security, and scalability.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical engineering, and more specifically, relates to a distributed real-time economic dispatch method and device for a microgrid system. Background Art

[0002] As an important form for accommodating renewable energy such as wind and light, microgrids have achieved rapid development worldwide. However, the randomness of new energy generation, electricity prices, loads, etc. poses challenges to the real-time operation of microgrids. At the same time, traditional centralized dispatch methods have disadvantages such as high communication and calculation costs, poor privacy, security, and scalability. Therefore, it is of great significance to study a distributed real-time economic dispatch method for a microgrid system.

[0003] As a stochastic optimization method, the approximate dynamic programming method has been widely applied in the real-time economic dispatch of microgrids. However, current research mostly only applies to centralized dispatch methods and cannot guarantee the distributed solution of the system. Some research attempts to apply the approximate dynamic programming method to distributed solutions. However, since the approximate dynamic programming method requires offline training and a large amount of communication and calculation costs are required in this process, the existing methods cannot guarantee the distributed nature during training, so the distributed real-time economic dispatch of microgrids cannot be guaranteed. Summary of the Invention

[0004] Aiming at the above defects or improvement requirements of the prior art, the present invention provides a distributed real-time economic dispatch method and device for a microgrid system. The purpose is to greatly reduce the difficulty of mapping state variables to value functions by using the method of state variable aggregation, solve the curse of dimensionality problem of the system state space, and in actual operation, be able to obtain information on random factors at future moments to assist system decision-making, and thus be able to quickly and accurately perform real-time optimal dispatch on the microgrid system.

[0005] To achieve the above object, according to one aspect of the present invention, there is provided a distributed real-time economic dispatch method for a microgrid system, including:

[0006] S1: Establish an intraday real-time dispatch model of the microgrid system according to the technical parameters of each component in the microgrid system; the objective function of the intraday real-time dispatch model is: minimizing the operating cost of the microgrid system in the total dispatch domain T time period;

[0007] S2: Reconstruct the intraday real-time economic dispatch model of the microgrid system into a distributed Markov decision process M t =<S t ,x t ,R t ,F t trans >; S tis the set of state variables at time t; x t is the set of decision variables at time t; R t is the set of random factors at time t; F t trans is the state transition equation indicating that the state variables at time t transfer according to the decision variables and random factors;

[0008] S3: Use a piecewise linear function to approximately describe the approximate optimal value function of S t from time t to the total scheduling domain T under the state; Based on the consensus-based distributed optimization algorithm, propose a distributed value function training method to train the piecewise linear function;

[0009] S4: Based on the random factor R of the microgrid system at time t t determine the state variable S at time t t , based on the trained piecewise linear function, sequentially and distributively solve the Markov decision process by minimizing the optimal operating cost function to obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t; Use the approximate optimal real-time economic dispatch decision to implement the dispatch of the microgrid system.

[0010] In one embodiment, the set of state variables of the microgrid system is including the spatial state variable and the time state variable λ t is the set of marginal costs of each generating unit in the distributed solution, is the power limit factor in the distributed solution, D t is the power mismatch of each generating unit in the distributed solution; is the output of the distributed generator g in the microgrid system at time t - Δt, is the state of charge of the energy storage b in the microgrid system at time t, P t w,a is the available output of the wind turbine w at time t, P t s,a is the available output of the photovoltaic unit s at time t, p t is the electricity price at time t, l t is the load at time t;

[0011] The set of decision variables of the microgrid system is x t ={P t g ; P t e ; P t b ; P t w ; Pt s} , P t b is the output of energy storage b in the microgrid system;

[0012] The set of random factors of the microgrid system includes the day-ahead prediction value set and the prediction error set where P t w,f is the day-ahead prediction value of wind turbine w at time t, P t s,f is the day-ahead prediction value of photovoltaic unit s at time t, p t f is the day-ahead prediction value of electricity price at time t, is the day-ahead prediction value of load at time t; is the prediction error of wind turbine w at time t, is the prediction error of photovoltaic unit s at time t, is the prediction error of electricity price at time t, is the prediction error of load at time t.

[0013] In one embodiment, the state transition equation of the microgrid system includes a spatial state transition equation and a temporal state transition equation:

[0014] Spatial state transition equation

[0015] Temporal state transition equation

[0016] where A is the weight matrix in distributed solution, ∈ λ , ∈ κ are feedback gain coefficients, idx is an index used to represent the nth element in the set; S t (idx) is the idxth element in the set S t .

[0017] In one embodiment, the S3 includes:

[0018] S31: Construct the optimal operation cost function of the microgrid system from time t to the total scheduling domain T time in the S t state C t (S t , x t ) is the operation cost at time t, is the state of charge of the energy storage at time t, is the approximation function, representing the optimal operation cost of the microgrid from time t + 1 to time T;

[0019] S32: Use a piecewise linear function to approximately replace the optimal operating cost function to obtain B is the set of energy storage in the microgrid; M is the total number of segments of the piecewise linear function; is the slope of the piecewise linear function; is the SOC value of energy storage b in segment m;

[0020] S33: Propose a distributed value function training method based on a consensus-based distributed optimization algorithm; use the value function training method to train the piecewise linear function to obtain an approximate optimal value function for auxiliary decision-making.

[0021] In one embodiment, the S33 includes:

[0022] S33-1. Initialize the piecewise linear function and set n = 1;

[0023] S33-2. According to the predicted information of wind power, photovoltaic power, electricity price, and load on the day-ahead, use the Monte Carlo method to generate a set of random factors of the microgrid system;

[0024] S33-3. Set k = Δk;

[0025] S33-4. Use a consensus-based distributed algorithm to solve the intra-day real-time economic dispatch problem of the microgrid. If the algorithm converges, output the decision of the microgrid; if the algorithm does not converge, set k = k + Δk to update the spatial state variables of the system and continue to use the consensus-based distributed algorithm to solve the intra-day real-time economic dispatch problem of the microgrid until the algorithm converges;

[0026] S33-5. According to the information of the energy storage module itself before and after the decision, distributively solve the sampling estimated value of the piecewise linear function;

[0027] S33-6. According to the sampling estimated value, through update the slope of the value function;

[0028] S33-7. Calculate the state S of the microgrid system at time t + Δt according to the state transition equation t+Δt ;

[0029] S33-8. Set t = t + Δt, and repeat steps S34 - S38 until t = T;

[0030] S33-9. Set n = n + 1, and repeat steps S32 - S39 until n = N; N is the preset number of iterations, thereby completing the training of the piecewise linear function.

[0031] In one of the embodiments, the distributed computing method for the sampling estimate value in S33-5 is as follows: Before making a decision, a small perturbation is added to the energy storage, and the microgrid is optimized again:

[0032] If the total power of the energy storage remains unchanged after optimization, then is used to calculate the sampling estimate value;

[0033] If the power of the energy storage changes after optimization, and the changed power of the energy storage is compensated by the distributed generating units, then is used to calculate the sampling estimate value;

[0034] If the changed power of the energy storage is compensated by the exchange power of the external power grid, then is used to calculate the sampling estimate value;

[0035] If the changed power of the energy storage is compensated by the exchange power of the wind power, then is used to calculate the sampling estimate value;

[0036] If the changed power of the energy storage is compensated by the exchange power of the photovoltaic power, then is used to calculate the sampling estimate value.

[0037] In one of the embodiments, S4 includes:

[0038] S41: Based on the random factor R of the microgrid system at time t t the state variable S at time t is determined t ;

[0039] S42: Based on the trained piecewise linear function, combined with the state variable S at time t t distributedly solve to obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t The dispatch of the microgrid system is implemented using the approximate optimal real-time economic dispatch decision.

[0040] In one of the embodiments, the objective function of the real-time economic dispatch model of the microgrid is:

[0041] where J represents the operating cost of the microgrid system in the total dispatch domain T time period; C t (S t , x t ) is the operating cost of the microgrid system at time t, G is the set of distributed generating units in the microgrid system, a g , b g , c g are the cost coefficients of the distributed generating unit g, P t gThe output of the distributed generator set g at time t; E is the set of external power grids in the microgrid system, p t is the electricity price at time t, P t e is the exchange power between the microgrid and the external power grid e at time t; W is the set of wind turbines in the microgrid system, P t w is the output of the wind turbine w in the microgrid system at time t, P t w,a is the theoretical output of the wind turbine in the microgrid system at time t, C w is the penalty cost for curtailed wind; S is the set of photovoltaic units in the microgrid system, P t s is the output of the photovoltaic unit s in the microgrid system at time t, P t s,a is the theoretical output of the photovoltaic unit s in the microgrid system at time t, C s is the penalty cost for curtailed light.

[0042] In one embodiment, the constraints of the intraday economic dispatch model of the microgrid system include: power balance constraint, ramp constraint of distributed generator sets, upper and lower limits constraint of the output of distributed generator sets, upper and lower limits constraint of the exchange power between the microgrid and the external power grid, upper and lower limits constraint of the output of wind power, upper and lower limits constraint of the output of photovoltaics, upper and lower limits constraint of the charge and discharge power of energy storage, charge state transfer constraint of energy storage, upper and lower limits constraint of energy storage capacity, and equality constraint of the initial and final state power of energy storage.

[0043] According to another aspect of the present invention, there is provided a distributed real-time economic dispatch device for a microgrid system, characterized in that it is used to execute the distributed real-time economic dispatch method of the microgrid system, including:

[0044] A modeling module, configured to establish an intraday real-time dispatch model of the microgrid system according to the technical parameters of each component in the microgrid system; the objective function of the intraday real-time dispatch model is: minimizing the operating cost of the microgrid system in the total dispatch domain T time period;

[0045] A reconstruction module, configured to reconstruct the intraday real-time economic dispatch model of the microgrid system into a distributed Markov decision process; is the set of state variables at time t; is the set of decision variables at time t; is the set of random factors at time t; is the state transition equation indicating that the state variables at time t are transferred according to the decision variables and random factors;

[0046] A training module, configured to approximately describe the approximate optimal value function from time t to the total dispatch domain T time period in the state by using a piecewise linear function; propose a distributed value function training method based on a consistency-based distributed optimization algorithm to train the piecewise linear function;

[0047] A scheduling module, configured to determine state variables at time t based on random factors of the microgrid system at time t, and based on the trained piecewise linear function, sequentially and distributively solve the Markov decision process by minimizing the optimal operation cost function, so as to obtain an approximate optimal real-time economic scheduling decision of the microgrid system at time t; and implement the scheduling of the microgrid system by using the approximate optimal real-time economic scheduling decision.

[0048] Generally speaking, compared with the prior art by the above technical solution conceived by the present invention, the following beneficial effects can be achieved:

[0049] 1. The present invention provides a distributed real-time economic scheduling method for a microgrid system. According to technical parameters of each component in the microgrid system, an intraday real-time scheduling model of the microgrid system is established; and the intraday real-time scheduling model of the microgrid system is reconstructed into a distributed Markov decision process; based on the piecewise linear function, an optimal operation cost function of the microgrid system from time t to the total scheduling domain T at state S t is constructed. Based on the state variables S t of the microgrid system at time t, combined with random factors of the microgrid system at time t, the Markov decision process is distributively solved by minimizing the optimal operation cost function, so as to obtain an approximate optimal real-time economic scheduling decision of the microgrid system at time t. The present invention greatly reduces the difficulty of mapping state variables to value functions by adopting the method of state variable aggregation, solves the curse of dimensionality problem of the system state space, and in actual operation, system operators can obtain information of random factors at future moments to assist system decision-making, and thus can quickly and accurately perform real-time optimal scheduling on the microgrid system.

[0050] 2. The distributed real-time economic scheduling method for a microgrid system provided by the present invention can combine a distributed optimization method based on consistency and approximate dynamic programming theory to propose a distributed piecewise linear value function training method, and simultaneously ensure the distributed property during day-ahead training and intraday real-time optimization.

[0051] 3. The real-time optimal scheduling method for a microgrid system provided by the present invention fully considers the influence of prediction errors of wind power, photovoltaics, electricity prices, and load day-ahead. Through the piecewise linear value function with excellent performance obtained by day-ahead training, the microgrid system can make an approximate optimal global approximate optimal real-time economic scheduling decision only according to random factors arriving at the current moment, and ensure the economy and safety of the microgrid system during intraday real-time operation.

[0052] 4. The distributed real-time economic dispatch method for a microgrid system provided by the present invention takes into account the impact of system prediction errors on intraday dispatch, while ensuring distributed communication among various components during the operation of the microgrid, reducing communication and computing costs in the actual system, and enhancing the privacy, security, and scalability of the microgrid system itself. The optimized real-time economic dispatch strategy for the microgrid system also ensures the economy and security of the microgrid system operation. Description of the Drawings

[0053] Figure 1 It is a flowchart of the distributed real-time economic dispatch method for the microgrid system provided by the present invention;

[0054] Figure 2 It is a microgrid structure diagram applied in the embodiment of the present invention;

[0055] Figure 3a It is a schematic diagram of the predicted curve and actual curve of wind power provided in the embodiment of the present invention;

[0056] Figure 3b It is a schematic diagram of the predicted curve and actual curve of photovoltaic provided in the embodiment of the present invention;

[0057] Figure 3c It is a schematic diagram of the predicted curve and actual curve of electricity price provided in the embodiment of the present invention;

[0058] Figure 3d It is a schematic diagram of the predicted curve and actual curve of load provided in the embodiment of the present invention;

[0059] Figure 4 It is a slope comparison diagram of the distributed real-time optimization dispatch method for the microgrid system provided by the present invention and the centralized ADP algorithm;

[0060] Figure 5 It is an effect comparison diagram of the distributed real-time optimization dispatch method for the microgrid system provided by the present invention and the distributed myopic algorithm. Detailed Embodiments

[0061] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0062] The present invention provides a distributed real-time economic dispatch method for a microgrid system, as Figure 1 shown, including the following steps:

[0063] S1. Establish an intraday real-time scheduling model for the microgrid system according to the technical parameters of each component in the microgrid system;

[0064] Specifically, the real-time economic scheduling model of the microgrid is:

[0065]

[0066] Among them, J represents the operating cost of the microgrid system in the total scheduling domain T time period; C t (S t , x t ) is the operating cost of the microgrid system at time t; G is the set of distributed generation units in the microgrid system, a g , b g , c g are the cost coefficients of distributed generation unit g, P t g is the output of distributed generation unit g at time t; E is the set of external power grids in the microgrid system, p t is the electricity price at time t, P t e is the exchange power between the microgrid and external power grid e at time t; W is the set of wind turbines in the microgrid system, P t w is the output of wind turbine w in the microgrid system at time t, P t w,a is the theoretical output of wind turbine in the microgrid system at time t, C w is the penalty cost for wind curtailment; S is the set of photovoltaic units in the microgrid system, P t s is the output of photovoltaic unit s in the microgrid system at time t, P t s,a is the theoretical output of photovoltaic unit s in the microgrid system at time t, C s is the penalty cost for PV curtailment.

[0067] The constraints of the intraday economic scheduling model of the microgrid system include: power balance constraint, ramp rate constraint of distributed generation units, upper and lower limits constraint of distributed generation units' output, upper and lower limits constraint of the exchange power between the microgrid system and the external power grid, upper and lower limits constraint of wind power output, upper and lower limits constraint of PV output, upper and lower limits constraint of energy storage charge and discharge power, energy storage state of charge transfer constraint, upper and lower limits constraint of energy storage capacity, and the constraint that the initial and final state electricity of the energy storage is equal.

[0068] S2. Reconstruct the intraday real-time economic scheduling model of the microgrid system into a distributed Markov decision process M t =<S t , x t , R t,F t trans >; S t is the set of state variables of the microgrid system at time t in the intra-day real-time economic dispatch model, S t includes spatial state variables and temporal state variables x t is the set of decision variables of the microgrid system at time t in the intra-day real-time economic dispatch model, corresponding to the decision of the microgrid system at time t; R t is the set of random factors of the microgrid system at time t; F t trans is the state transition equation representing the transition of state variables at time t according to decision variables and random factors, including spatial state transition equation and temporal state transition equation;

[0069] Specifically, the set of state variables of the microgrid system is includes spatial state variables and temporal state variables where λ t is the set of marginal costs of each generating unit in distributed solution, is the power limit factor in distributed solution, D t is the power mismatch of each generating unit in distributed solution; is the output of distributed generator g in the microgrid system at time t - Δt, is the state of charge of energy storage b in the microgrid system at time t, P t w,a is the available output of wind turbine w at time t, P t s,a is the available output of photovoltaic unit s at time t, p t is the electricity price at time t, l t is the load at time t.

[0070] The set of decision variables of the microgrid system is x t ={P t g ; P t e ; P t b ; P t w ; P t s}, where P t b is the output of energy storage b in the microgrid system.

[0071] The set of random factors of the microgrid system includes the set of day-ahead prediction values and the set of prediction errors Among them, P t w,f is the day-ahead prediction value of wind turbine w at time t, and P t s,f is the day-ahead prediction value of photovoltaic unit s at time t, is the day-ahead prediction value of electricity price at time t, is the day-ahead prediction value of load at time t; is the prediction error of wind turbine w at time t, is the prediction error of photovoltaic unit s at time t, is the prediction error of electricity price at time t, is the prediction error of load at time t.

[0072] The state transition equation of the microgrid system includes a spatial state transition equation and a time state transition equation:

[0073] Spatial state transition equation

[0074] Time state transition equation

[0075] Among them, A is the weight matrix in distributed solution, ∈ λ , ∈ κ is the feedback gain coefficient, and idx is the index used to represent the nth element in the set; S t (idx) is the idxth element in the set S t in.

[0076] S33. Construct the optimal operating cost function of the microgrid system from time t to the total scheduling domain T time under the S t state and use the method of piecewise linear function approximation to replace the original approximation function Combined with the distributed optimization algorithm based on consensus, a distributed value function training method is proposed to obtain an approximate optimal value function for auxiliary decision-making;

[0077] S4. Based on the random factors of the microgrid system at time t, determine the state variable S of the microgrid system at time t t , combined with the trained approximate optimal value function, by minimizing the optimal operating cost function, sequentially and distributedly solve the Markov decision process to obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t.

[0078] To more clearly illustrate a distributed real-time economic dispatch method for a microgrid system proposed by the present invention, the following will be described in detail with specific embodiments:

[0079] Taking a 6-node microgrid system as an example for analysis, as Figure 2As shown in the figure. The system has 1 distributed generator set (denoted as g), 1 node connected to the external power grid (denoted as e), 1 wind turbine generator set (denoted as w), 1 load node (denoted as l), 1 photovoltaic generator set (denoted as s), and 1 energy storage unit (denoted as b).

[0080] First, collect the technical parameters of each component in the microgrid system; among them, the components of the microgrid system include distributed generator sets, external power grids, wind turbine generator sets, loads, photovoltaic generator sets, and energy storage units. The technical parameters of each component specifically include:

[0081] 1) The number N of microgrid nodes bus , wind power P w,f , photovoltaic power P s,f , electricity price p f , load l f of the day-ahead prediction value; 2) The number of lines N branch of the microgrid, the node numbers at both ends of the line; 3) The node number where the distributed generator set is located, the upper and lower limits of the output of the distributed generator set and the maximum up and down ramp rates and as well as the cost coefficients a g , b g and c g ; 4) The node number where the external power grid unit is located, the upper limit of the exchange power between the microgrid and the external power grid 5) The nodes where the wind turbine generator set, load, and photovoltaic generator set are located 6) The node where the energy storage unit is located, the upper limit of the energy storage power The upper and lower limits of the capacity of the energy storage unit and as well as the initial power of the energy storage unit

[0082] In this embodiment, the parameters of the system components are shown in Table 1:

[0083] Table 1 Parameters

[0084] Device <![CDATA[P max (kW)]]> <![CDATA[P min (kW)]]> Ramp Rate (kW / h) <![CDATA[a g ($ / kWh^2)]]> <![CDATA[b g ($ / kWh)]]> <![CDATA[c g ($)]]> Distributed Generation 80 20 50 0.0005 0.0397 0.4 External Power Grid 20 \ \ \ \ \ Wind Power 40 0 \ \ \ \ Photovoltaic 35 0 \ \ \ \ Energy Storage 15 0 \ \ \ \

[0085] The distributed real-time economic dispatch method of the microgrid system considering randomness in this embodiment specifically includes the following steps:

[0086] S1. According to the technical parameters of each component in the microgrid system, establish an intraday real-time dispatch model of the microgrid system: The objective function is:

[0087]

[0088] The corresponding constraint conditions are specifically:

[0089] Power balance constraint: P t g +P t e +P t w +P t s +P t b =l t ;

[0090] Ramp constraint of distributed generation units:

[0091] Upper and lower limits constraint of distributed generation unit output:

[0092] Upper and lower limits constraint of the exchanged power between the microgrid system and the external power grid:

[0093] Upper and lower limits constraint of wind power output: 0 ≤ P t w ≤ P t w,a ;

[0094] Upper and lower limits constraint of PV output: 0 ≤ P t s ≤ P t s,a ;

[0095] Upper and lower limits constraint of energy storage charge and discharge power:

[0096] Energy storage state of charge transfer constraint:

[0097] Upper and lower limits constraint of energy storage capacity:

[0098] Energy storage initial and final state of charge equality constraint:

[0099] S2. Reconstruct the intra-day real-time economic dispatch model of the microgrid system into a distributed Markov decision process M t =<S t ,x t ,R t ,F t trans >; S t is the set of state variables of the microgrid system at time t in the intra-day real-time economic dispatch model, S t includes spatial state variables and temporal state variables x tLet \(R\) be the set of decision variables of the microgrid system at time \(t\) in the intra-day real-time economic dispatch model, corresponding to the decision of the microgrid system at time \(t\). t Let \(F\) be the set of random factors of the microgrid system at time \(t\). t trans Let \(S\) be the state transition equation representing the transition of the state variable according to the decision variable and the random factor at time \(t\), including the spatial state transition equation and the time state transition equation.

[0100] Specifically, the state variable reflects the current state of the microgrid system. In this embodiment, the state variable is divided into a spatial state variable and a time state variable. The spatial state variable acts on the single-period distributed optimization solution, and the time state variable acts on the multi-period optimization solution. The set of spatial state variables of the microgrid at time \(t\) is The set of time state variables of the microgrid at time \(t\) is

[0101] Specifically, the decision variable reflects the decision of the microgrid system at the current moment. In this embodiment, the set of decision variables is \(x\) t =\(\{P t g ; P t e ; P t b ; P t w ; P t s \}\);

[0102] Specifically, the random factor reflects the randomness of the microgrid system. In this embodiment, the random factor includes the set of day-ahead prediction values and the set of prediction errors

[0103] Specifically, the state transition equation reflects the change of the state of the microgrid system. In this embodiment, the state transition equation is divided into a spatial state transition equation and a time state transition equation:

[0104]

[0105]

[0106] S3. Construct the optimal operating cost function of the microgrid system from time \(t\) to the total dispatching domain \(T\) at state \(S\) t and use the method of piecewise linear function approximation to replace the original approximation function Combined with the distributed optimization algorithm based on consistency, a distributed value function training method is proposed to obtain the approximate optimal value function for auxiliary decision-making;

[0107] ​Specifically, the steps for training the distributed value function in this microgrid are as follows:

[0108] S331. Initialize the piecewise linear function and set n = 1;

[0109] S332. According to the predicted information of wind power, photovoltaics, electricity price, and load for the day-ahead, use the Monte Carlo method to generate a set of random factors for the microgrid system;

[0110] S333. Set k = Δk;

[0111] S334. Use the distributed algorithm based on consensus to solve the intra-day real-time economic dispatch problem of the microgrid. If the algorithm converges, output the decision of the microgrid; if the algorithm does not converge, set k = k + Δk to update the spatial state variables of the system and continue to use the distributed algorithm based on consensus to solve the intra-day real-time economic dispatch problem of the microgrid until the algorithm converges.

[0112] S335. According to the information of the energy storage module in the microgrid system before and after the decision, distributively solve the sampling estimation value of the piecewise linear function. The calculated sampling estimation value is only related to the information of the energy storage module. The distributed calculation method of the sampling estimation value is as follows, which is divided into five cases. Add a small perturbation to the energy storage before the decision and re-optimize the microgrid. If the total power of the energy storage remains unchanged after optimization, use formula (a) to calculate the sampling estimation value; if the power of the energy storage changes after optimization and the changed power of the energy storage is compensated by the distributed generator set, use formula (b) to calculate the sampling estimation value. If the changed power of the energy storage is compensated by the exchange power of the external power grid, use formula (c) to calculate the sampling estimation value. If the changed power of the energy storage is compensated by the exchange power of wind power, use formula (d) to calculate the sampling estimation value. If the changed power of the energy storage is compensated by the exchange power of photovoltaics, use formula (e) to calculate the sampling estimation value.

[0113]

[0114]

[0115]

[0116]

[0117]

[0118] S336. According to the calculated sampling estimation value, Update the slope of the value function and use the leveling algorithm to ensure the convexity of the piecewise linear function.

[0119] S337. Calculate the state S of the microgrid system at time t+Δt according to the state transition equation t+Δt ;

[0120] S338. Let t = t+Δt, and repeat steps S334 - S338 until t = T;

[0121] S339. Let n = n+1, and repeat steps S332 - S339 until n = N; N is the preset number of iterations.

[0122] S4. Based on the random factors of the microgrid system at time t, determine the state variable S of the microgrid system at time t t , and combine with the trained approximate optimal value function. By minimizing the optimal operation cost function, sequentially and distributively solve the Markov decision process to obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t.

[0123] Specifically, the method for distributively solving the Markov decision process of the distributed real-time economic dispatch method of the microgrid system considering randomness includes: based on the trained piecewise linear function, and distributively solve in combination with the state variable of the microgrid system at time t to obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t

[0124] To further illustrate the effectiveness of the distributed real-time dispatch method of the microgrid system considering randomness provided by the present invention, under the same randomness scenario, the distributed real-time dispatch method of the microgrid provided by the present invention (denoted as D-ADP), the centralized ADP algorithm (denoted as C-ADP), and the distributed myopic algorithm (denoted as D-Myopic) are respectively used to optimize the dispatch of the microgrid system. The randomness scenario is as Figure 3a - Figure 3d shown, where Figure 3a is the wind power randomness scenario, Figure 3b is the photovoltaic randomness scenario, Figure 3c is the electricity price randomness scenario, Figure 3d is the load randomness scenario. The comparison between the proposed D-ADP method and the centralized dispatch result is as Figure 4 shown. It can be seen that the absolute value of the error between the proposed distributed value function update strategy and the centralized update strategy is less than 1e-8, which proves the effectiveness of the proposed distributed value function update strategy. The comparison with the distributed myopia is as Figure 5 shown. The average optimization error of the real-time economic dispatch of the microgrid provided by the present invention is 0.67%, while the optimization error of the distributed myopic algorithm is 12.64%. Therefore, the real-time optimization dispatch method of the microgrid system considering randomness provided by the present invention has a better effect than the myopic algorithm, and when applied to actual examples, it can greatly improve the safety and economy of the operation of the microgrid system.

[0125] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A distributed real-time economic dispatch method for a microgrid system, characterized in that, including: S1: Establish an intraday real-time scheduling model of the microgrid system according to the technical parameters of each component in the microgrid system; the objective function of the intraday real-time scheduling model is: minimizing the operating cost of the microgrid system in the total scheduling domain T time period; the intraday real-time scheduling model is: Among them, J represents the operating cost of the microgrid system in the total dispatching domain T time period; C t (S t , x t ) is the operating cost of the microgrid system at time t; G is the set of distributed generation units in the microgrid system, a g , b g , c g are the cost coefficients of the distributed generation unit g, P t g is the output of the distributed generation unit g at time t; E is the set of external power grids in the microgrid system, p t is the electricity price at time t, P t e is the exchange power between the microgrid and the external power grid e at time t; W is the set of wind turbines in the microgrid system, P t w is the output of the wind turbine w in the microgrid system at time t, P t w,a is the theoretical output of the wind turbine in the microgrid system at time t, C w is the penalty cost for wind curtailment; S is the set of photovoltaic units in the microgrid system, P t s is the output of the photovoltaic unit s in the microgrid system at time t, P t s,a is the theoretical output of the photovoltaic unit s in the microgrid system at time t, C s is the penalty cost for PV curtailment; S2: Reconstruct the intraday real-time scheduling model of the microgrid system into a distributed Markov decision process M t =<S t ,x t ,R t ,F t trans >; S t is the set of state variables at time t; x t is the set of decision variables at time t; R t is the set of random factors at time t; F t trans is the state transition equation indicating the transition of the state variable at time t according to the decision variable and the random factor; S3: Use a piecewise linear function to approximately describe S t The approximate optimal value function from time t to the total scheduling domain T time in the state; Based on the consensus-based distributed optimization algorithm, a distributed value function training method is proposed to train the piecewise linear function; S4: Based on the random factor R of the microgrid system at time t t Determine the state variable S at time t t , based on the trained piecewise linear function, solve the Markov decision process sequentially and distributively by minimizing the optimal operating cost function to obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t; use the approximate optimal real-time economic dispatch decision to implement the dispatch of the microgrid system; The S3 includes: S31: Construct the optimal operation cost function of the microgrid system from time t to the total scheduling domain T at state S t The optimal operation cost function from time t to the total scheduling domain T at state S C t (S t , x t ) is the operation cost at time t, is the state of charge of the energy storage at time t, is the approximate value function, representing the optimal operation cost of the microgrid from time t + 1 to T; S32: Use a piecewise linear function to approximately replace the optimal operating cost function to obtain B is the set of energy storage in the microgrid; M is the total number of segments of the piecewise linear function; is the slope of the piecewise linear function; is the SOC value of energy storage b in segment m; S33: Propose a distributed value function training method based on the consensus-based distributed optimization algorithm; use the value function training method to train the piecewise linear function to obtain an approximate optimal value function for auxiliary decision-making.

2. The distributed real-time economic scheduling method of the microgrid system according to claim 1, characterized in that The set of state variables of the microgrid system is including spatial state variables and temporal state variables λ t is the set of marginal costs of each generating unit in distributed solution, is the power limit factor in distributed solution, D t is the power mismatch of each generating unit in distributed solution; is the output of distributed generator g in the microgrid system at time t - Δt, is the state of charge of energy storage b in the microgrid system at time t, P t w,a is the available output of wind turbine w at time t, P t s,a is the available output of photovoltaic unit s at time t, p t is the electricity price at time t, l t is the load at time t; The set of decision variables of the microgrid system is x t ={P t g ; P t e ； P t b ； P t w ； P t s}, P t b is the output of energy storage b in the microgrid system; The set of random factors of the microgrid system includes the day-ahead prediction value set and the prediction error set where, P t w,f is the day-ahead prediction value of wind turbine w at time t, P t s,f is the day-ahead prediction value of photovoltaic unit s at time t, p t f is the day-ahead prediction value of electricity price at time t, is the day-ahead prediction value of load at time t; is the prediction error of wind turbine w at time t, is the prediction error of photovoltaic unit s at time t, is the prediction error of electricity price at time t, is the prediction error of load at time t.

3. The distributed real-time economic dispatch method of the microgrid system according to claim 2, characterized in that The state transition equation of the microgrid system includes a spatial state transition equation and a time state transition equation: Spatial state transition equation Time state transition equation Among them, A is the weight matrix in distributed solution, ∈ λ , ∈ κ is the feedback gain coefficient, idx is the index, used to represent which element in the set; S t (idx) is the idx-th element in the set S t .

4. The distributed real-time economic dispatch method of the microgrid system according to claim 1, characterized in that The S33 includes: S33-1: Initialize the piecewise linear function and set n = 1; S33-2: According to the predicted information of wind power, photovoltaic power, electricity price, and load on the previous day, use the Monte Carlo method to generate a set of random factors of the microgrid system; S33-3: Set k = Δk; S33-4: Use the consensus-based distributed algorithm to solve the intraday real-time economic scheduling problem of the microgrid. If the algorithm converges, output the decision of the microgrid; if the algorithm does not converge, set k = k + Δk to update the spatial state variables of the system and continue to use the consensus-based distributed algorithm to solve the intraday real-time economic scheduling problem of the microgrid until the algorithm converges; S33-5: According to the information of the energy storage module before and after the decision, distributedly solve the sampling estimated value of the piecewise linear function; S33-6. Based on the sampled estimated value, update the slope of the value function by ; S33-7. Calculate the state S of the microgrid system at time t+Δt according to the state transition equation t+Δt ; S33-8: Set t = t + Δt, and repeat steps S34 - S38 until t = T; S33-9: Set n = n + 1, and repeat steps S32 - S39 until n = N; N is a preset number of iterations, thus completing the training of the piecewise linear function.

5. The distributed real-time economic dispatch method of the microgrid system according to claim 4, characterized in that, The distributed calculation method of the sampling estimated value in S33-5 is as follows: Add a small perturbation to the energy storage before the decision and re-optimize the microgrid: If the total power of the energy storage after optimization remains unchanged, then use to calculate the sampling estimate value; If the power of the energy storage changes after optimization and the changed power of the energy storage is compensated by the distributed generator sets, then use to calculate the sampling estimate value; If the power change of the energy storage is compensated by the exchange power of the external power grid, then calculate the sampling estimated value; If the power change of energy storage is compensated by the exchange power of wind power, then calculate the sampling estimate value; If the power change of energy storage is compensated by the exchange power of PV, then calculate the sampling estimate value.

6. The distributed real-time economic dispatch method of the microgrid system according to claim 1, characterized in that The S4 includes: S41: Determine the state variable S at time t based on the random factor R of the microgrid system at time t t t ;​ S42: Based on the trained piecewise linear function, combined with the state variable S at time t t Distributed solution Obtain the approximate optimal real-time economic dispatch decision of the microgrid system at time t Implement the dispatch of the microgrid system by using the approximate optimal real-time economic dispatch decision 7. The distributed real-time economic dispatch method of the microgrid system according to claim 6, wherein, The constraints of the intraday economic scheduling model of the microgrid system include: power balance constraint, ramp constraint of distributed generators, upper and lower limits of distributed generator output, upper and lower limits of the exchange power between the microgrid system and the external power grid, upper and lower limits of wind power output, upper and lower limits of photovoltaic power output, upper and lower limits of energy storage charge and discharge power, energy storage state of charge transfer constraint, upper and lower limits of energy storage capacity, and the constraint that the initial and final state electricity of the energy storage is equal.

8. A distributed real-time economic dispatch device for a microgrid system, characterized in that, For executing the distributed real-time economic scheduling method of the microgrid system according to any one of claims 1 - 7, including: A modeling module, configured to establish an intraday real-time scheduling model of the microgrid system according to the technical parameters of each component in the microgrid system; the objective function of the intraday real-time scheduling model is: minimizing the operating cost of the microgrid system in the total scheduling domain T time period; A reconstruction module, which is used to reconstruct the intra-day real-time scheduling model of the microgrid system into a distributed Markov decision process; $\mathbf{x}_t$ is the set of state variables at time $t$; $\mathbf{u}_t$ is the set of decision variables at time $t$; $\mathbf{w}_t$ is the set of random factors at time $t$; $f(\mathbf{x}_{t + 1},\mathbf{x}_t,\mathbf{u}_t,\mathbf{w}_t)$ is the state transition equation representing the transition of the state variable at time $t$ according to the decision variable and the random factor; A training module, which is used to approximately describe the approximate optimal value function from time $t$ to the total scheduling domain $T$ at state $s$ by using a piecewise linear function; propose a distributed value function training method based on a consensus-based distributed optimization algorithm to train the piecewise linear function; A scheduling module, which is used to determine the state variable at time $t$ based on the random factor of the microgrid system at time $t$, and based on the trained piecewise linear function, sequentially and distributively solve the Markov decision process by minimizing the optimal operating cost function to obtain the approximate optimal real-time economic scheduling decision of the microgrid system at time $t$; use the approximate optimal real-time economic scheduling decision to implement the scheduling of the microgrid system.

Images