Power distribution optimization scheduling method and system suitable for commercial complex microgrid

Abstract

The invention discloses a power distribution optimization scheduling method and system suitable for a commercial complex microgrid. According to the invention, a modular layered structure is introduced in modeling, and the whole system is divided into a main power supply scheduling layer, a renewable output prediction response layer and an energy storage flexible adjustment layer; and a mathematical optimization + reinforcement learning cooperation mechanism is adopted in solving, a deterministic part is accurately optimized by using an MILP model, an uncertain part is subjected to online generation of a strategy by using a reinforcement learning algorithm, and rapid iterative updating and feedback control of a scheduling result are realized. And meanwhile, through a boundary coupling coordination algorithm, synchronous updating of state variables among the sub-models and global feasibility of a scheduling solution are guaranteed, so that the response speed, the solution quality and the system robustness of the whole scheduling system are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the cross field of micro-grid optimization scheduling and intelligent control, and particularly relates to a power distribution optimization scheduling method and system suitable for commercial complex micro-grid. BACKGROUND

[0002] With the development of renewable energy technology and the promotion of urban energy structure transformation, more and more commercial complexes, smart parks and cultural tourism areas are deploying localized micro-grid systems to improve energy autonomy and risk resistance. Such urban load centers are usually equipped with photovoltaic power generation, energy storage systems and emergency backup power sources to support local independent operation in the case of peak-valley price difference, power limitation, maintenance or failure. Commercial complex micro-grid has the characteristics of high load intensity, strong demand regularity but severe fluctuation, and frequent grid-connected / off-grid switching. At the same time, its energy system structure is complex, and the running environment changes frequently, which puts higher requirements on the real-time performance, economy and robustness of the scheduling system.

[0003] The existing micro-grid optimization scheduling method has obvious shortcomings in dealing with the characteristics of urban commercial complex micro-grid system such as "high load uncertainty, high volatility, multi-energy collaborative complexity", and it is difficult to generate a high-quality scheduling scheme while meeting the supply and demand balance, operation cost control and efficient use of renewable energy. There are mainly three technical bottlenecks: First, the optimization model is highly complex and difficult to adapt to the dynamic re-optimization requirements in the rolling time domain. Most current scheduling models use centralized scheduling architecture, which needs to consider multiple power types such as photovoltaic PV, wind power WT, diesel engine DG, energy storage system ESS, as well as load forecasting, power balance, state constraints and other variables, and covers time sequence coupling, multi-energy flow collaboration and network constraints. Such models are often built as mixed integer linear programming (MILP) or nonlinear programming (NLP) forms. In the context of multi-period, multi-device and multi-objective, the number of decision variables and constraint conditions increases rapidly, resulting in a significant increase in solving time, especially in the rolling optimization (Rolling Horizon Optimization, RHO) framework, which is difficult to respond to changing conditions in real time, reducing the engineering practicability of the model.

[0004] Secondly, the lack of effective hierarchical modeling and coupling-decoupling strategies easily leads to "dimension disaster" and "computational bottleneck" problems. In the typical commercial complex independent microgrid, there are complex operation logic and nonlinear constraints between various distributed energy resources (DERs), such as start-stop control of diesel or natural gas combined heat and power equipment, state of charge (SOC) limitation of energy storage system, uncontrollability of photovoltaic output, etc. The traditional single-layer centralized modeling method fails to fully consider the heterogeneity between different energy units in terms of control accuracy, response delay and scheduling period, resulting in high overall coupling degree of the system, rapid expansion of optimization problem dimension, and thus seriously affecting the feasibility and convergence efficiency of the global solution. In addition, the current method generally lacks a boundary coupling processing mechanism for key nodes in scheduling, making it difficult to effectively coordinate the synchronous update of state variables between energy storage, power supply and load side in actual urban operation scenarios, further limiting the landing and promotion ability of the model in commercial complex microgrid.

[0005] Thirdly, the algorithm design lacks a collaborative control mechanism for real-time and robustness. In systems with significant uncertainty, existing scheduling strategies rely on historical data for static modeling, making it difficult to effectively deal with renewable output forecast error or load fluctuation. On the one hand, although precise algorithms based on MILP or dynamic programming (DP) have high accuracy, they respond slowly to changes in system state and cannot output a feasible solution within the scheduling window. On the other hand, although heuristic algorithms based on particle swarm optimization (PSO), genetic algorithm (GA), etc. can provide fast approximate solutions, they lack constraint feasibility guarantee mechanisms, resulting in poor result stability and low operability, and cannot meet the safety and reliability requirements of system scheduling. SUMMARY

[0006] The application aims to provide a power distribution optimization scheduling method and system suitable for a commercial complex microgrid. The application introduces a modular hierarchical structure in modeling, divides the overall system into a main power source scheduling layer (Diesel Generator and ESS Dispatch Layer), a renewable power prediction response layer (Renewable Forecast and Matching Layer), and an energy storage flexibility adjustment layer (Energy Storage Flexibility Layer); in solving, a collaborative mechanism of "mathematical optimization + reinforcement learning (RL)" is adopted, the MILP model is used for accurate optimization of the deterministic part, and the reinforcement learning algorithm is used to generate strategies online for the uncertain part, realizing the rapid iteration and feedback control of the scheduling results. At the same time, through the boundary coupling coordination algorithm, the synchronous update of the state variables between the sub-models and the global feasibility of the scheduling solution are guaranteed, thereby improving the response speed, solution quality and system robustness of the entire scheduling system.

[0007] To achieve the object of the application, the technical solutions provided by the application are as follows: First aspect The application provides a power distribution optimization scheduling method suitable for a commercial complex microgrid, comprising the following steps: Step S1: establishing a hierarchical decoupled microgrid scheduling model; Step S2: decomposing the microgrid scheduling model into three sub-models, and using a hybrid optimization algorithm to solve in parallel, and using a boundary coupling coordination algorithm to check the global feasibility and synchronization of the solution, to obtain a scheduling plan; wherein the three sub-models include a renewable energy prediction response layer, a main power source scheduling layer, and an energy storage flexibility control layer; Step S3: based on a rolling horizon optimization mechanism, performing closed-loop execution and dynamic feedback update on the scheduling plan generated in step S2, and outputting a final control instruction set after coordination and verification.

[0008] Second aspect Corresponding to the above method, the application further provides a power distribution optimization scheduling system suitable for a commercial complex microgrid, comprising the following units: a model establishing unit, a scheduling plan obtaining unit, and a final control instruction set output unit; The model establishing unit is used to establish a hierarchical decoupled microgrid scheduling model; The scheduling plan obtaining unit is configured to decompose the micro-grid scheduling model into three-layer sub-models, and solve the three-layer sub-models in parallel by using a hybrid optimization algorithm, and perform global feasibility checking and synchronization on the solving results by using a boundary coupling coordination algorithm to obtain the scheduling plan. The final control instruction set output unit is configured to perform closed-loop execution and dynamic feedback updating on the scheduling plan generated by the scheduling plan obtaining unit based on a rolling horizon optimization mechanism, and output the final control instruction set after coordination and checking.

[0009] Compared with the prior art, the beneficial effects of the present application are as follows: The present application provides a power distribution optimization scheduling solving method suitable for commercial complex micro-grid, aiming at the characteristics of high load uncertainty and high fluctuation of energy of commercial complex micro-grid, a hierarchical decoupling scheduling model is constructed, and a hybrid solving framework combining reinforcement learning and optimization algorithm is introduced. The method divides the whole problem into three sub-layers of renewable energy prediction and response, main power supply scheduling and energy storage regulation, and solves them by using mixed integer programming, probability modeling and reinforcement learning respectively, which effectively reduces the computational complexity. The results of each sub-module are integrated by a global coordination mechanism, and real-time scheduling updating is realized in the rolling horizon, which improves the economy, robustness and response speed of the system, and is suitable for various urban commercial complex micro-grid system scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A power distribution optimization scheduling method suitable for commercial complex micro-grid is provided for the embodiments of the present application. Figure 2 A solving time comparison diagram in a comparison test between the method of the present application and the traditional method is shown. Figure 3 A supply-demand imbalance rate comparison diagram in a comparison test between the method of the present application and the traditional method is shown. Figure 4 A target function fluctuation range comparison diagram in a comparison test between the method of the present application and the traditional method is shown. Figure 5 A comparison diagram of energy storage SOC stability in a comparison test between the method of the present application and the traditional method is shown. DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0012] It should be noted that this application is particularly applicable to the optimization problems of independently operating microgrids in urban commercial complexes in terms of power supply coordination, load forecasting, and energy storage dispatch.

[0013] like Figure 1 As shown, this invention provides a power distribution optimization scheduling method suitable for microgrids in commercial complexes, comprising the following steps: Step S1: Establish a hierarchical decoupled microgrid scheduling model; Specifically, step S1 includes the following: Step S1.1: Define the decision variables for the microgrid dispatch model, including the following: Diesel generator at time ; output power; Energy storage systems at all times The charging power; Energy storage systems at all times The discharge power; The state of charge of the energy storage system at time t; Step S1.2: Define the parameters for each level of the microgrid dispatch model, including the following: Total number of time periods in the scheduling cycle; : The fuel cost function of a diesel generator at time t; : Charging efficiency of energy storage systems; Discharge efficiency of the energy storage system; The predicted output power of the photovoltaic system at time t; The predicted output power of the wind power system at time t; Load demand at time t; The total network loss of the microgrid system at time t; Step S1.3: Establish the objective function of the microgrid scheduling model, including the following: ; In the formula: Fuel costs of diesel generators; Energy storage lifespan loss cost; The cost of penalties for abandoning wind and solar power; Costs resulting from supply and demand imbalance.

[0014] Step S1.4: Establish the constraints for the microgrid dispatch model, including the following: (1.4.1) Power balance constraints: ; (1.4.2) Diesel engine output limit: ; In the formula: : lower limit of diesel generator output power; : upper limit of diesel generator output power; (1.4.3) Dynamic update of energy storage SOC: ; In the formula: : state of charge of energy storage system at time t+1; is a time-varying quantity; (1.4.4) Energy storage safety interval constraint: ; In the formula: : lower limit of energy storage system SOC; : upper limit of energy storage system SOC; (1.4.5) Charge-discharge exclusion constraint: .

[0015] Step S2: decompose the microgrid scheduling model into three layers of sub-models, and solve in parallel using a hybrid optimization algorithm, and use boundary coupling coordination algorithm for global feasibility checking and synchronization of the solution results to obtain the scheduling plan; wherein the three layers of sub-models include a renewable energy prediction response layer, a main power source scheduling layer and an energy storage flexible control layer; The renewable energy prediction response layer is an ultra-short-term prediction based on LSTM, which is a forecasting model providing high-precision data support for other layers. The model formula of the renewable energy prediction response layer is as follows: ; ; ; ; In the formula: : input vector at time t; : hidden state at time t; : hidden state at time t-1; , , : forgetting gate, input gate and output gate, respectively; : cell state at time t; : cell state at time t-1; : forgetting gate weight matrix; : forgetting gate bias vector; : candidate cell state at time t; W o: weight matrix of output layer; b o : bias vector of output layer; σ: Sigmoid activation function : Hadamard product

[0016] The solving step of the model formula of the renewable energy prediction response layer includes the following: Input: real-time collected historical time series data Calculation: input the historical time series data into the trained LSTM network to perform forward propagation calculation Output (solving result): photovoltaic ultra-short-term point prediction value for the next 15 minutes to 4 hours and wind power ultra-short-term point prediction value and confidence interval of the prediction value

[0017] The main power dispatching layer is a long-term economic dispatching model based on MILP. According to the output of the prediction response layer, the main power dispatching layer formulates a micro-grid dispatching plan based on the economy of the long-term operation of the system. The model formula of the main power dispatching layer includes the following: (2.1) Model decision variable: : output power of diesel generator at time t : charging power of energy storage system at time t : discharging power of energy storage system at time t : start-stop state of diesel generator at time t : state of charge of energy storage system at time t (2.2) Model objective function, as follows: ; ; In the formula, T is the time period : diesel fuel cost : diesel generator operation and maintenance cost coefficient : diesel generator start-up cost : start-up flag of diesel generator at time t : power shortage penalty cost coefficient : power shortage at time t (2.3) Model constraint conditions, including the following: (2.3.1) Power balance constraint, formula as follows: ; Wherein: is the load prediction value is the photovoltaic power prediction value ​For wind power prediction output; (2.3.2) Diesel generator constraints Upper and lower output constraints: ; In the formula: : Start-stop state of diesel generator at time t, 0 or 1; Ramp rate constraints: ; ; In the formula: , : Upper and lower limits of diesel generator ramp rate, respectively; : Output power of diesel generator at time period -1; Minimum start-stop time constraints: ; ; In the formula: , : Minimum start and stop times, respectively; is the continuous running time of diesel generator at time t-1, is the continuous stop time of diesel generator at time t-1; is the start-stop state at time t-1, 0 or 1; Start flag constraints: ; (2.3.3) Energy storage system constraints Charge and discharge power limits: ; ; In the formula, is the maximum charging power; is the charging start flag at time t; is the maximum discharging power; is the discharging start flag at time t; Charge-discharge mutual exclusion constraints: ; SOC safety range constraints: ; (2.3.4) Power shortage definition .

[0018] In addition, the model formula solving step of the main power dispatch layer is as follows: Input: 24-hour future load prediction curve, wind and light prediction curve, current energy storage SOC initial value, equipment parameters and cost coefficient; Modeling: build the above objective function and constraint condition into a MILP model; Solving: call commercial solvers such as Gurobi and CPLEX for solving; Output (solving result): diesel generator all-day start-stop plan and output plan curve, energy storage system all-day plan charge-discharge curve, and energy storage system all-day reference SOC trajectory .

[0019] Among them, the energy storage flexible control layer is a real-time optimization control based on DDPG, which models the scheduling problem as a Markov decision process (MDP) and learns the optimal strategy to real-time suppress fluctuations. The model formula of the energy storage flexible control layer includes the following: State space (State, ): ; In the formula, is the upper layer plan information; Action space (Action, ): ; In the formula, is the adjustment amount relative to the upper layer plan; Reward function (Reward, ): ; Among them: , , , are the weight coefficients of each reward item; is the power value at time t; State transition: determined by the energy storage SOC dynamic equation in the environment: ; In the formula, is the rated capacity of the energy storage.

[0020] Among them, the solving step of the model formula of the energy storage flexible control layer is as follows: Input: current state obtained from the environment ; Policy evaluation: input state into the trained DDPG algorithm Actor network ; Action selection: Actor network outputs the current optimal action where is noise; Execution and observation: execute the action , the environment moves to a new state , and gets a reward ; Output (solution result): energy storage real-time power control instruction in the next control cycle .

[0021] It should be noted that the boundary coupling coordination algorithm is to ensure the consistency and global feasibility of each layer solution.

[0022] The coordination algorithm is as follows: Coordination logic: Get the actual current SOC of the energy storage, the plan of the main power supply layer , and the prediction after the action execution of the DDPG layer ; Check the conflict, as follows: check whether ; wherein, is the safety margin; If a conflict occurs, generate a coordination signal: To the MILP layer: add constraints wherein, is the given correction amount; Trigger its re-solution; To the DDPG layer: modify its reward function, increase the penalty weight of the SOC out-of-limit item ; Solution steps: Input: Collect all layer inputs, outputs, and system actual measurement values.

[0023] Calculation: execute the coordination algorithm.

[0024] Output (solution result): A globally feasible set of predictive control instructions, issued to physical devices.

[0025] When the set of predictive control instructions conflicts, output re-optimization instructions and new boundary conditions.

[0026] Step S3: Based on the rolling horizon optimization mechanism, perform closed-loop execution and dynamic feedback update on the scheduling plan generated in step S2, and output the final control instruction set after coordination and verification.

[0027] The step S3 specifically includes the following: Step S3.1: Start the rolling optimization window at a fixed period, collect the latest actual state data of the micro-grid system (load, wind and solar output, SOC), and trigger the renewable energy prediction response layer to update the short-term prediction; Step S3.2: Re-execute the solving process of step S2 with the latest actual state data of the micro-grid system as input: for the main power supply scheduling layer, re-solve the MILP-based long-term economic dispatching model with the current state as the initial value to generate an updated day-ahead plan; for the energy storage flexible control layer, call the DDPG policy network to generate real-time control instructions according to the updated micro-grid system state; Step S3.3: Call the boundary coupling coordination algorithm to check the consistency of the output results of each layer after re-solution If a conflict is found, immediately backtrack to the coordination process for adjustment until a feasible solution is obtained; Step S3.4: Output the final control instruction set after coordination and verification, including diesel generator short-term output instructions, real-time charging and discharging power instructions of the energy storage system, and wind and solar consumption strategies, forming a closed-loop control of optimization-execution-feedback-reoptimization.

[0028] In addition, the present application also includes the step of verifying the final control instruction set through a simulation platform, as follows: Verify the final control instruction set through a simulation platform (Matlab / Simulink) to ensure power balance and equipment safety. If the wind and solar fluctuations exceed the prediction range, trigger the reinforcement learning module to re-generate the energy storage strategy. In extreme scenarios (such as typhoon weather), switch to a robust mode to prioritize critical load power supply.

[0029] In order to fully evaluate the performance of the method in solving speed, optimization effect and stability in actual operation, a set of comparative experiments are designed, and traditional mixed integer linear programming (MILP), ROME method based on robust optimization and the method proposed in the present application are selected for comparative analysis. The experiments are carried out in terms of optimization problem size, system robustness under renewable energy fluctuations, target function stability and energy storage system operation safety, etc., to comprehensively verify the performance of the method in multiple dimensions.

[0030] Figure 2 The solving time performance of the three methods under different optimization problem sizes is shown. As the problem size gradually increases from small to large, the solving time of MILP and ROME method increases exponentially and quickly approaches or reaches the set maximum solving time (about 10 4Compared with the above methods, the application also provides a power distribution optimization scheduling system suitable for a commercial complex micro-grid, which comprises the following units: a model establishing unit, a scheduling plan obtaining unit and a final control instruction set output unit.

[0031] Figure 3 For supply-demand imbalance rate comparison (wind and light fluctuation greater than 30%), the supply-demand imbalance rates of the three methods under the uncertainty levels of ±0%, ±15% and ±30% are compared in the scenario considering the high fluctuation of renewable energy. The results show that the supply-demand imbalance rate of the application method is significantly lower than that of MILP and ROME under all fluctuation levels, which shows stronger robustness and system coordination ability.

[0032] Figure 4 For target function fluctuation range comparison, the figure reflects the stability of the target function of the solutions obtained by the three methods. The target function of MILP and ROME fluctuates greatly in different experiments, while the fluctuation range of the results of the application is the smallest, which shows that it has stronger optimization stability and can maintain higher running consistency in the face of disturbance or uncertainty.

[0033] As shown in Figure 5 , the SOC stability comparison compares the daily variation curves of the state of charge (SOC) of the energy storage system under the three methods. The SOC curve under the control of the application fluctuates more gently and always maintains in the safe operation interval, while the SOC under the MILP and ROME schemes fluctuates more greatly and even fluctuates outside the critical interval, which shows that the application method is more reliable and has practical application value in system dynamic control.

[0034] In addition, corresponding to the above methods, the application also provides a power distribution optimization scheduling system suitable for a commercial complex micro-grid, which comprises the following units: a model establishing unit, a scheduling plan obtaining unit and a final control instruction set output unit. The model establishing unit is used to establish a hierarchical decoupled micro-grid scheduling model. The scheduling plan obtaining unit is used to decompose the micro-grid scheduling model into three-layer sub-models, and solve them in parallel by using a hybrid optimization algorithm, and perform global feasibility checking and synchronization on the solving results by using a boundary coupling coordination algorithm to obtain a scheduling plan; wherein the three-layer sub-models comprise a renewable energy prediction response layer, a main power supply scheduling layer and an energy storage flexible control layer. The final control instruction set output unit is used to perform closed-loop execution and dynamic feedback update on the scheduling plan generated by the scheduling plan obtaining unit based on a rolling horizon optimization mechanism, and output a final control instruction set after coordination and verification.

[0035] Specifically, the model building unit is used to perform the following: Step S1.1: Define the decision variables for the microgrid dispatch model, including the following: Diesel generator at time ; output power; Energy storage systems at all times The charging power; Energy storage systems at all times The discharge power; The state of charge of the energy storage system at time t; Step S1.2: Define the parameters for each level of the microgrid dispatch model, including the following: Total number of time periods in the scheduling cycle; : The fuel cost function of a diesel generator at time t; : Charging efficiency of energy storage systems; : Discharge efficiency of the energy storage system; The predicted output power of the photovoltaic system at time t; The predicted output power of the wind power system at time t; Load demand at time t; The total network loss of the microgrid system at time t; Step S1.3: Establish the objective function of the microgrid scheduling model, including the following: ; In the formula: Fuel costs of diesel generators; Energy storage lifespan loss cost; The cost of penalties for abandoning wind and solar power; Costs resulting from supply and demand imbalance.

[0036] Step S1.4: Establish the constraints for the microgrid dispatch model, including the following: (1.4.1) Power balance constraints: ; (1.4.2) Diesel engine output limit: ; In the formula: Lower limit of diesel generator output power; : Maximum output power of diesel generator; (1.4.3) Dynamic updates of energy storage SOC: ; In the formula: The state of charge of the energy storage system at time t+1; It is a quantity that changes over time; (1.4.4) Energy storage safety range constraints: ; In the formula: : Lower limit of SOC for energy storage systems; : Upper limit of SOC for energy storage systems; (1.4.5) Charge-discharge mutual exclusion constraint: .

[0037] The renewable energy forecasting response layer is an ultra-short-term forecasting method based on LSTM. The model formula for the renewable energy forecasting response layer is as follows: ; ; ; ; In the formula: : The input vector at time t; The hidden state at time t; The hidden state at time t-1; , , These are the forget gate, input gate, and output gate, respectively. : The state of t cells at any given time; : Cell state at time t-1; Forget gate weight matrix; Forget gate bias vector; : Candidate cell state at time t; W o : Weight matrix of the output layer; b o : The bias vector of the output layer; σ: The sigmoid activation function; Hadamard product.

[0038] The main power supply scheduling layer is a long-term economic scheduling model based on MILP, and the model formula of the main power supply scheduling layer includes the following: (2.1) Model decision variables: Diesel generator at time ; output power; Energy storage systems at all times The charging power; Energy storage systems at all times The discharge power; : The start-stop state of the diesel generator at time t; SOC(t): The state of charge of the energy storage system at time t; (2.2) Model objective function, as follows: ; wherein: T is the time period; : diesel fuel cost; : diesel generator operation and maintenance cost coefficient; : diesel generator start-up cost; : diesel generator start-up flag at time t; : power shortage penalty cost coefficient; : power shortage at time t; (2.3) Model constraints, including the following: (2.3.1) Power balance constraint, formula as follows: ; wherein: is the load forecast value; is the photovoltaic power forecast output; is the wind power forecast output; (2.3.2) Diesel generator constraints Output upper and lower limit constraint: ; wherein: : start-stop state of diesel generator at time t, 0 or 1; Ramp rate constraint: ; ; wherein: , : upper and lower limits of diesel generator ramp rate, respectively; : output power of diesel generator at time period -1; Minimum start-stop time constraint: ; ; wherein: , : minimum start-up and stop time, respectively; is the continuous running time of diesel generator at time t-1, is the continuous stop time of diesel generator at time t-1; is the start-stop state at time t-1, 0 or 1; Start-up flag constraint: ; (2.3.3) Energy storage system constraints Charge and discharge power limits: ; ; wherein, is the maximum charge power; is the charge start flag at time t; is the maximum discharge power; is the discharge start flag at time t; Charge and discharge mutual exclusion constraints: ; SOC safety range constraints: ; (2.3.4) Power deficit definition .

[0039] It should be noted that the above-mentioned embodiments are only used for illustrating and describing the present application, and are not intended to limit the present application to the described embodiments. Furthermore, those skilled in the art can understand that the present application is not limited to the above-mentioned embodiments, and more kinds of variations and modifications can be made according to the teaching of the present application, and these variations and modifications all fall within the scope of the present application.

Claims

1. A power distribution optimization scheduling method suitable for microgrids in commercial complexes, characterized in that, Includes the following steps: Step S1: Establish a hierarchical decoupled microgrid scheduling model; Step S2: Decompose the microgrid dispatch model into three sub-models and solve them in parallel using a hybrid optimization algorithm. Use a boundary coupling coordination algorithm to perform global feasibility verification and synchronization on the solution results to obtain the dispatch plan. The three sub-models include a renewable energy prediction and response layer, a main power dispatch layer, and an energy storage flexible control layer. Step S3: Based on the rolling time-domain optimization mechanism, perform closed-loop execution and dynamic feedback updates on the scheduling plan generated in step S2, and output the final control instruction set after coordination and verification.

2. The power distribution optimization and dispatching method for microgrids in commercial complexes according to claim 1, characterized in that, Step S1 specifically includes the following: Step S1.1: Define the decision variables for the microgrid dispatch model, including the following: Diesel generator at time ; output power; Energy storage systems at all times The charging power; Energy storage systems at all times The discharge power; The state of charge of the energy storage system at time t; Step S1.2: Define the parameters for each level of the microgrid dispatch model, including the following: Total number of time periods in the scheduling cycle; : The fuel cost function of a diesel generator at time t; : Charging efficiency of energy storage systems; Discharge efficiency of the energy storage system; The predicted output power of the photovoltaic system at time t; The predicted output power of the wind power system at time t; Load demand at time t; The total network loss of the microgrid system at time t; Step S1.3: Establish the objective function of the microgrid scheduling model, including the following: ； In the formula: Fuel costs of diesel generators; Energy storage lifespan loss cost; The cost of penalties for abandoning wind and solar power; Supply and demand imbalance costs; Step S1.4: Establish the constraints for the microgrid dispatch model, including the following: (1.4.1) Power balance constraints: ； (1.4.2) Diesel engine output limit: ； In the formula: Lower limit of diesel generator output power; : Maximum output power of diesel generator; (1.4.3) Dynamic updates of energy storage SOC: ； In the formula: The state of charge of the energy storage system at time t+1; It is a quantity that changes over time; (1.4.4) Energy storage safety range constraints: ； In the formula: : Lower limit of SOC for energy storage systems; : Upper limit of SOC for energy storage systems; (1.4.5) Charge-discharge mutual exclusion constraint: 。 3. The power distribution optimization and scheduling method for microgrids in commercial complexes according to claim 2, characterized in that, In step S2, the renewable energy prediction response layer is an ultra-short-term prediction based on LSTM. The model formula for the renewable energy prediction response layer is as follows: ； ； ； ； In the formula: : The input vector at time t; The hidden state at time t; The hidden state at time t-1; , , These are the forget gate, input gate, and output gate, respectively. : The state of t cells at any given time; : Cell state at time t-1; Forget gate weight matrix; Forget gate bias vector; : Candidate cell state at time t; W o : Weight matrix of the output layer; b o : The bias vector of the output layer; σ: Sigmoid activation function; Hadamard product.

4. The power distribution optimization and scheduling method for microgrids in commercial complexes according to claim 3, characterized in that, In step S2, the main power supply scheduling layer is a long-term economic scheduling model based on MILP, and the model formula of the main power supply scheduling layer includes the following: (2.1) Model decision variables: Diesel generator at time ; output power; Energy storage systems at all times The charging power; Energy storage systems at all times The discharge power; : The start-stop state of the diesel generator at time t; SOC(t): The state of charge of the energy storage system at time t; (2.2) The objective function of the model is as follows: ； In the formula: T represents the time period; Diesel fuel cost; Diesel generator operation and maintenance cost coefficient; Diesel generator start-up costs; The starting flag of the diesel generator at time t; Power deficit penalty cost coefficient; The power deficit at time t; (2.3) Model constraints, including the following: (2.3.1) Power balance constraint, the formula is as follows: ； in: This is the load forecast value; For photovoltaic power prediction; This is the output for wind force prediction. (2.3.2) Diesel generator constraints Output upper and lower limit constraints: ； In the formula: : The start / stop status of the diesel generator at time t, 0 or 1; Slope rate constraint: ； ； In the formula: , These are the upper and lower limits of the gradeability of the diesel generator; Diesel generator during the period -1 output power; Minimum start-stop time constraints: ； ； In the formula: , These represent the minimum start and stop times, respectively. At time t-1, the diesel generator has been running continuously for [time period]. The diesel generator has been continuously stopped for the duration of time t-1. The start / stop status at time t-1 is 0 or 1; Startup flag constraint: ； (2.3.3) Constraints of Energy Storage Systems Charge and discharge power limits: ； ； In the formula, This represents the maximum charging power. The charging start flag is set at time t; This represents the maximum discharge power. This is the discharge start flag at time t; Charge-discharge mutual exclusion constraint: ； SOC security range constraints: ； (2.3.4) Definition of power deficit ； 。 5. The power distribution optimization and scheduling method for microgrids in commercial complexes according to claim 4, characterized in that, In step S2, the energy storage flexible control layer is a real-time optimization control based on DDPG, and the model formula of the energy storage flexible control layer includes the following: State space ): ； In the formula, Information for higher-level planning; Action space ): ； In the formula, This refers to the adjustment amount relative to the upper-level plan; Reward function ): ； in: , , , These are the weighting coefficients for each reward item; It is the power value at time t; State transition: determined by the dynamic equations of energy storage SOC in the environment: ； In the formula: This refers to the rated capacity of the energy storage.

6. The power distribution optimization and scheduling method for microgrids in commercial complexes according to claim 5, characterized in that, Step S3 specifically includes the following: Step S3.1: Start the rolling optimization window at a fixed period, collect the latest actual state data of the microgrid system, and trigger the renewable energy prediction response layer to update the ultra-short-term forecast; Step S3.2: Using the latest actual state data of the microgrid system as input, re-execute the solution process of step S2: For the main power dispatch layer, re-solve the long-term economic dispatch model based on MILP with the current state as the initial value to generate an updated day-ahead plan; for the energy storage flexible control layer, call the DDPG strategy network to generate real-time control commands according to the updated microgrid system state. Step S3.3: Call the boundary coupling coordination algorithm to verify the consistency of the output results of each layer after resolving. sex; If a conflict is found, the process is immediately reverted back to the coordination process for readjustment until a feasible solution is found. Step S3.4: Output the final control command set after coordination and verification, including the recent output of the diesel generator. Commands, real-time charging and discharging power commands of the energy storage system, and wind and solar energy consumption strategies form a closed-loop control system of optimization, execution, feedback, and re-optimization.

7. A power distribution optimization and dispatching system suitable for microgrids in commercial complexes, characterized in that, It includes the following units: model building unit, scheduling plan acquisition unit, and final control instruction set output unit; The model building unit is used to build a hierarchical and decoupled microgrid scheduling model; The scheduling plan obtaining unit is used to decompose the microgrid scheduling model into three sub-models, and solve them in parallel using a hybrid optimization algorithm. The solution results are then used to perform global feasibility verification and synchronization using a boundary coupling coordination algorithm to obtain the scheduling plan. The three sub-models include a renewable energy prediction and response layer, a main power supply scheduling layer, and an energy storage flexible control layer. The final control instruction set output unit is used to perform closed-loop execution and dynamic feedback update of the scheduling plan generated by the scheduling plan obtaining unit based on the rolling time domain optimization mechanism, and output the final control instruction set after coordination and verification.

8. A power distribution optimization and dispatching system suitable for microgrids in commercial complexes according to claim 7, characterized in that, The model building unit is specifically used to perform the following: Step S1.1: Define the decision variables for the microgrid dispatch model, including the following: Diesel generator at time ; output power; Energy storage systems at all times The charging power; Energy storage systems at all times The discharge power; The state of charge of the energy storage system at time t; Step S1.2: Define the parameters for each level of the microgrid dispatch model, including the following: Total number of time periods in the scheduling cycle; : The fuel cost function of a diesel generator at time t; : Charging efficiency of energy storage systems; Discharge efficiency of the energy storage system; The predicted output power of the photovoltaic system at time t; The predicted output power of the wind power system at time t; Load demand at time t; The total network loss of the microgrid system at time t; Step S1.3: Establish the objective function of the microgrid scheduling model, including the following: ； In the formula: Fuel costs of diesel generators; Energy storage lifespan loss cost; The cost of penalties for abandoning wind and solar power; Supply and demand imbalance costs; Step S1.4: Establish the constraints for the microgrid dispatch model, including the following: (1.4.1) Power balance constraints: ； (1.4.2) Diesel engine output limit: ； In the formula: Lower limit of diesel generator output power; : Maximum output power of diesel generator; (1.4.3) Dynamic updates of energy storage SOC: ； In the formula: The state of charge of the energy storage system at time t+1; It is a quantity that changes over time; (1.4.4) Energy storage safety range constraints: ； In the formula: : Lower limit of SOC for energy storage systems; : Upper limit of SOC for energy storage systems; (1.4.5) Charge-discharge mutual exclusion constraint: 。 9. A power distribution optimization and dispatching system suitable for microgrids in commercial complexes according to claim 8, characterized in that, The renewable energy forecasting response layer is an ultra-short-term forecast based on LSTM. The model formula for the renewable energy forecasting response layer is as follows: ； ； ； ； In the formula: : The input vector at time t; The hidden state at time t; The hidden state at time t-1; , , These are the forget gate, input gate, and output gate, respectively. : The state of t cells at any given time; : Cell state at time t-1; Forget gate weight matrix; Forget gate bias vector; : Candidate cell state at time t; W o : Weight matrix of the output layer; b o : The bias vector of the output layer; σ: Sigmoid activation function; Hadamard product.

10. A power distribution optimization and dispatching system suitable for microgrids in commercial complexes according to claim 9, characterized in that, The main power supply scheduling layer is based on the MILP long-term economic scheduling model, and the model formula of the main power supply scheduling layer includes the following: (2.1) Model decision variables: Diesel generator at time ; output power; Energy storage systems at all times The charging power; Energy storage systems at all times The discharge power; : The start-stop state of the diesel generator at time t; SOC(t): The state of charge of the energy storage system at time t; (2.2) The objective function of the model is as follows: ； In the formula: T represents the time period; Diesel fuel cost; Diesel generator operation and maintenance cost coefficient; Diesel generator start-up costs; The starting flag of the diesel generator at time t; Power deficit penalty cost coefficient; The power deficit at time t; (2.3) Model constraints, including the following: (2.3.1) Power balance constraint, the formula is as follows: ； in: This is the load forecast value; For photovoltaic power prediction; This is the output for wind force prediction. (2.3.2) Diesel generator constraints Output upper and lower limit constraints: ； In the formula: : The start / stop status of the diesel generator at time t, 0 or 1; Slope rate constraint: ； ； In the formula: , These are the upper and lower limits of the gradeability of the diesel generator; Diesel generator during the period -1 output power; Minimum start-stop time constraints: ； ； In the formula: , These represent the minimum start and stop times, respectively. At time t-1, the diesel generator has been running continuously for [time period]. The diesel generator has been continuously stopped for the duration of time t-1. The start / stop status at time t-1 is 0 or 1; Startup flag constraint: ； (2.3.3) Constraints of Energy Storage Systems Charge and discharge power limits: ； ； In the formula, This represents the maximum charging power. The charging start flag is set at time t; This represents the maximum discharge power. This is the discharge start flag at time t; Charge-discharge mutual exclusion constraint: ； SOC security range constraints: ； (2.3.4) Definition of power deficit ； 。

Images