Wind and light storage micro-grid black-start control method based on model predictive control

Abstract

The invention provides a wind and light storage micro-grid black-start control method based on model prediction control, and belongs to the field of electric power system operation control, and the method comprises the following steps: building a wind and light storage micro-grid structure model, and respectively adjusting the wind power plant unit number, the photovoltaic power plant unit scale and the discharge power of an energy storage system, managing the output levels of the wind power generation subsystem and the photovoltaic power generation subsystem; wind and light power generation power and energy storage charge and discharge power of the wind and light storage micro-grid are predicted through a double-layer control framework, the energy storage charge state is adjusted, and the double-layer control framework comprises a global optimization layer and a power coordination layer. The double-layer control architecture reduces the power fluctuation of the wind and light storage micro-grid, and prevents the energy storage from being overcharged and overdischarged.

Description

Technical Field

[0001] This invention belongs to the field of power system operation control, specifically relating to a black start control method for wind-solar-storage microgrids based on model predictive control. Background Technology

[0002] With the continuous expansion of the power grid and the increasing penetration rate of new energy sources, the black start capability of microgrids has become the last line of defense for ensuring power system security and is crucial for improving the efficiency of power grid recovery after disasters. As a locally autonomous power system, a microgrid typically includes distributed generation, energy storage, and loads, and can operate independently during main grid failures. Wind, solar, and energy storage power stations, with their advantages of abundant local resources, flexible start-up and shutdown, and environmental friendliness, are increasingly regarded as an important supplement or even alternative to traditional hydropower or diesel black start power sources, playing an increasingly important role in the rapid recovery of regional power grids after a complete outage. However, with the increasing penetration rate of new energy sources, the fluctuation factors faced by the system in the isolated operation mode of microgrids are becoming increasingly complex. The intermittency of wind and solar power output, the abrupt changes in load input, and the cumulative effect of prediction errors can cause significant power deviations in a short period of time. If these deviations are not effectively suppressed, they will not only increase the regulation pressure on the energy storage system but may also lead to system frequency exceeding limits or voltage instability, ultimately triggering cascading failures, making power balance during the black start phase a severe challenge.

[0003] While the traditional "wind and solar maximum power point tracking + energy storage difference compensation" strategy is structurally simple, it is difficult to effectively cope with the random fluctuations in wind and solar power output during actual black start processes. It also lacks fine management of the charging and discharging process of the energy storage system, which can easily lead to the energy storage unit being in a deep charging and discharging state frequently, accelerating the aging of energy storage equipment, and even causing overcharging or over-discharging of energy storage, thus threatening the smooth progress of the black start process. Therefore, traditional open-loop simulation and static optimization methods that only focus on large disturbance transient processes or long-term stability performance are no longer suitable for the comprehensive performance evaluation and real-time optimization requirements of modern power systems for wind, solar and energy storage black start processes in terms of rapid response, operational safety, and minimum loss.

[0004] The patent document with publication number CN120896262A discloses a method and device for adjusting output power. It determines the target operating parameters of the power station based on the operating data of different buses and equipment in the power network, monitors the output power of the energy storage equipment cluster port and the real-time active and reactive power output of energy routing, and then dynamically adjusts the output power of the power transmission station. This method alleviates power fluctuations and improves grid stability when the new energy power generation system is connected to the grid. However, this method does not design a hierarchical collaborative control architecture for the black start scenario of wind-solar-storage microgrids, and it is difficult to cope with the dynamic optimization requirements of random fluctuations in wind and solar power output and balanced management of energy storage charge state during the black start process. Summary of the Invention

[0005] To address the problems in the prior art, this invention proposes a black-start control method for wind-solar-storage microgrids based on model predictive control. It adopts a two-layer control architecture of global optimization layer and power coordination layer to reduce power fluctuations in wind-solar-storage microgrids, prevent overcharging and over-discharging of energy storage, and improve the reliability of black-start of wind-solar-storage microgrids.

[0006] The technical solution of the present invention is as follows: This invention proposes a black-start control method for wind-solar-storage microgrids based on model predictive control, comprising the following steps: A microgrid structure model for wind, solar and energy storage is established. By adjusting the number of wind turbines in operation, the scale of photovoltaic power station input units, and the discharge power of the energy storage system, the output levels of the wind power generation system and the photovoltaic power generation system are managed. The wind power generation, photovoltaic power generation and energy storage charging and discharging power of the wind-solar-storage microgrid are predicted by a two-layer control architecture, and the energy storage state of charge is adjusted. The two-layer control architecture includes a global optimization layer and a power coordination layer. The global optimization layer employs a model prediction algorithm based on reinforcement learning, continuously adjusting the weights of the predictive control objective function according to the system's operating state, and generating the optimal control model through the reward function. The power coordination layer adopts a model predictive control-based power coordination strategy for black start of wind, solar and energy storage microgrids. It establishes a closed-loop control framework of predictive model, rolling optimization and feedback correction to control the power during the black start process of wind, solar and energy storage microgrids.

[0007] Furthermore, the specific output levels of the wind power generation system and the photovoltaic power generation system are as follows: The regulation and control of wind farms and photovoltaic power plants is described as follows: ; in, N pv ( k ) is the target time k The number of photovoltaic units; N pv ( k +1) represents the target time. k +1 to the number of photovoltaic units; Δ N pv ( k (time) k At the time k +1 represents the change in the number of photovoltaic units; N w ( k ) is the target time k The number of wind turbines; N w (k +1) represents the target time. k +1 number of fans; Δ N w (k) represents time (k) k At the time k +1 represents the change in the number of wind turbines; The process of power output change in wind farms and photovoltaic power plants is as follows: ; in, P pv ( k ) is the target time k Measured output power of photovoltaic power station; P pv ( k +1) represents the target time. k +1 Measured output power of photovoltaic power station; P pvn ( k The output power of the photovoltaic unit is derived from ultra-short-term power prediction. P w ( k ) is the target time k Measured output power of the wind farm; P w ( k +1 represents the target time. k +1 Measured output power of wind farm; P wn ( k The output power of the wind turbine is derived from the ultra-short-term power prediction.

[0008] Furthermore, the energy conversion process of the energy storage system can be represented as follows: ; in, E B ( k ) is the target time k Real-time energy storage capacity; E ( k +1) represents the target time. k +1 real-time energy storage capacity; η The conversion efficiency of energy storage; Δ T B This is the conversion factor from power to electrical quantity; P B ( k ) is the target time k Energy storage capacity.

[0009] Furthermore, the reinforcement learning-based model prediction algorithm is specifically as follows: A reference trajectory is generated based on the current state of the wind-solar-storage microgrid system. y s ( k Based on machine learning models, future dynamics are predicted to obtain energy storage forecast values. y m ( k+j ); Optimize the first step control amount of the preset scrolling. u ( k The input is the rolling optimization module, which solves for the optimal control sequence based on the objective function, applies it to the controlled object, and outputs the actual value of energy storage. y ( k ); Sensors collect actual energy storage values y ( k The input is fed into the neural network to calculate the state-action value function Q(s|a), and an evaluation signal is given based on the reward function of the rolling optimization module. The control parameters or strategies of the model predictive control are dynamically adjusted, and the control decision is fed back into the correction loop. y ( k Through feedback correction process and y m ( k+j Compare and generate y r ( k + j The output is corrected to correct the prediction model error, and rolling optimization and feedback correction are repeated.

[0010] Furthermore, in the global optimization layer, the grid frequency deviation, frequency change rate, and charge state distribution balance index are used as the system's state-space equations as follows: ; Where, Δ f t This represents the current frequency deviation. Entropy is the rate of change of frequency. SOC t () is an index of the uniformity of the state of charge distribution; The weights of the model predictive control objective function are continuously adjusted based on the system's operating state to balance the relative importance of different objectives. Its action space state expression is as follows: ; in, α t For time t Frequency deviation suppression weights; β t For time tMinimize the weight of energy storage losses; γ t For time t The weights for balancing the state of charge; After executing the model predictive control flow, the system provides numerical feedback. The reward function guides the DRL to learn a multi-objective cooperative strategy, and its expression is: ; in, w 1. w 2 and w 3 is the normalization coefficient; P ess For frequency modulation output; The normalization coefficient is used to balance the priorities of various objectives and to penalize frequency deviation, energy storage loss and state of charge imbalance.

[0011] Furthermore, the state-space model equation of the power coordination layer prediction model is as follows: ; The state variable is represented as: ; in, P B ( k )for k The charging and discharging power of energy stored at all times; The control variable is represented as: ; External disturbance r ( k )= P load , P load For load power; System output y ( k )= P B , P B The charging and discharging power for energy storage; Matrices A, B1, B2, and C are shown below: ; ; ; ; D 1 and D 2 is set as a zero matrix.

[0012] Furthermore, the prediction model for a wind power generation system or a photovoltaic power generation system can be obtained by shielding another subsystem and setting its matrix components to zero, and then substituting it into the state-space model equation.

[0013] Furthermore, the cost function for the rolling optimization is as follows: ; Where M is the time step; The cost function, which includes optimization metrics such as frequency deviation, energy storage loss, and state-of-charge balance, is shown below: ; The cost function for limiting the frequency of switching is shown below: ; Local optimization of each energy storage unit in distributed energy storage i In the prediction time domain T The system executes the initial control command to perform rolling updates and solves the equation. The equation expression is: ; in, Indicates the first i One energy storage unit from t +1 to t + T Minimize the frequency modulation output sequence within the predicted time domain; t + k | t Indicates time t For future moments t + k The predicted value; Considering the power and state of charge constraints, the following conditions are met: ; ; in ，P rated,i For the first i The rated power of each energy storage unit; For the first i Minimum permissible state of charge for each energy storage unit; For the first i The maximum permissible charge of each energy storage unit.

[0014] Furthermore, an entropy function is introduced to quantify the uniformity index of the state of charge distribution. The specific definition of the entropy function is: ; ; in, SOCi For the first i The state of charge of each energy storage unit; N This represents the number of energy storage units.

[0015] Furthermore, through a real-time feedback correction mechanism, the actual operating state of the system is used as the initial condition for the next optimization cycle to compensate for the impact of wind and solar power prediction errors.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention employs a two-layer intelligent control architecture, enabling multi-objective dynamic optimization. The upper layer, based on a reinforcement learning-based model predictive control algorithm, can perceive key state indicators such as system frequency deviation, rate of frequency change, and energy storage charge state distribution balance in real time, and dynamically adjust the multi-objective weight coefficients of the lower-layer model predictive controller. This design allows the system to dynamically prioritize multiple conflicting objectives such as frequency stability, minimizing energy storage losses, and achieving cluster charge state balance, thus realizing multi-objective collaborative optimization during black start and overcoming the limitations of traditional control methods in single-objective optimization. This invention integrates prediction and feedback mechanisms to enhance system robustness. By establishing a state-space prediction model for the wind-solar-storage system and combining rolling optimization and real-time feedback correction, a complete closed-loop control framework is formed. In each control cycle, the system predicts future operating trends based on the current state, solves for the optimal control sequence, and corrects prediction errors in real time through the feedback mechanism. This design effectively compensates for the impact of uncertainties in wind and solar power prediction, significantly reduces power fluctuations, prevents overcharging and over-discharging of energy storage, and reduces the configuration requirements for energy storage capacity, enabling the system to exhibit stronger adaptability and reliability in complex black-start environments. Attached Figure Description

[0017] Figure 1 The topology of the wind-solar-storage microgrid structure model; Figure 2 This is a schematic diagram of the algorithm principle for the global optimization layer; Figure 3 This is a control principle diagram for the power coordination layer. Detailed Implementation

[0018] To make the features and beneficial effects of the present invention more apparent and understandable, the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0019] Example 1 This embodiment provides a black-start control method for wind-solar-storage microgrids based on model predictive control, including the following steps: First, a microgrid structure model of wind, solar, and energy storage is established, with the specific topology as follows: Figure 1As shown, this topology model integrates wind power generation systems and photovoltaic power generation systems, connected to the main grid at the point of common coupling (PCC) through the microgrid's main isolation equipment. Internally, it includes renewable energy generation units composed of small wind turbines and photovoltaic cells, as well as local loads divided into sensitive and non-sensitive loads. The energy manager works in conjunction with the microgrid dispatch center to be responsible for the system's power balance and optimized scheduling. The protection coordinator quickly detects and isolates faults, such as short circuits and overloads, ensuring that faults are limited to a minimum, thereby guaranteeing continuous power supply to sensitive loads and overall system safety. The power flow controller adjusts the output of photovoltaic units and wind turbines, as well as the charging and discharging of the energy storage system, to achieve precise management of active and reactive power flow within the microgrid, maintaining system frequency and voltage stability. The separators, i.e., circuit breakers, perform physical opening and closing operations under the command of the protection coordinator or according to preset conditions, enabling seamless switching between grid-connected and islanded modes of the microgrid, and controlling the switching of non-sensitive loads. The collaboration of the protection coordinator, power flow controller, and separators further enhances the system's control flexibility and operational stability.

[0020] Preferably, the power flow controller is implemented by an energy manager.

[0021] Wind power generation systems and photovoltaic power generation systems are naturally complementary in terms of resource characteristics, reducing dependence on a single energy source. This allows the system to relax the power and capacity requirements for energy storage under the same load demand, thus reducing the system's dependence on energy storage.

[0022] To enhance the power regulation flexibility of wind-solar-storage combined power generation systems, this invention employs a control method suitable for black start operations. By adjusting the number of operating wind turbines, the scale of photovoltaic power station units, and the discharge power of the energy storage system, precise management of the output levels of each subsystem is achieved. This ensures that the system can provide a continuous and stable power supply to the load during black start, while maintaining the state of charge (SOC) of the energy storage units within a safe and efficient operating range. The control process for the photovoltaic power station and wind farm is described as follows: ; in, N pv ( k ) is the target time k The number of photovoltaic units; N pv ( k +1) represents the target time. k +1 to the number of photovoltaic units; Δ N pv ( k (time) k At the time k +1 represents the change in the number of photovoltaic units;N w ( k ) is the target time k The number of wind turbines; N w ( k +1) represents the target time. k +1 number of fans; Δ N w (k) represents time (k) k At the time k +1 represents the change in the number of wind turbines; The process of power output change for photovoltaic power plants and wind farms is as follows: ; in, P pv ( k ) is the target time k Measured output power of photovoltaic power station; P pv ( k +1) represents the target time. k +1 Measured output power of photovoltaic power station; P pvn ( k The output power of the photovoltaic unit is derived from ultra-short-term power prediction. P w ( k ) is the target time k Measured output power of the wind farm; P w ( k +1 represents the target time. k +1 Measured output power of wind farm; P wn ( k The output power of the wind turbine is derived from ultra-short-term power prediction. The state of charge (SBC) of energy storage is the ratio of the current charge of the stored energy to its rated capacity. The transformation process of the stored energy charge is represented as follows: ; in, E B ( k ) is the target time k Real-time energy storage capacity; E ( k +1) represents the target time. k +1 real-time energy storage capacity; η The conversion efficiency of energy storage; Δ T B This is the conversion factor from power to electrical quantity; P B (k ) is the target time k Energy storage capacity.

[0023] This invention employs a two-layer control architecture consisting of a global optimization layer and a power coordination layer. The global optimization layer utilizes a reinforcement learning-based model prediction algorithm, the principle of which is as follows: Figure 2 As shown, this algorithm is based on a closed-loop feedback mechanism and integrates the rolling optimization of model predictive control (MPC) and the adaptive decision-making capabilities of reinforcement learning (RL). Its working principle is as follows: A reference trajectory is generated based on the current state of the wind-solar-storage microgrid system. y s ( k Based on machine learning models, future dynamics are predicted to obtain energy storage forecast values. y m ( k+j ); Optimize the first step control amount of the preset scrolling. u ( k The input is the rolling optimization module, which solves for the optimal control sequence based on the objective function, applies it to the controlled object, and outputs the actual value of energy storage. y ( k ); Sensors collect actual energy storage values y ( k The input is fed into the neural network to calculate the state-action value function Q(s|a), and an evaluation signal is given based on the reward function of the rolling optimization module. The control parameters or strategies of the model predictive control are dynamically adjusted, and the control decision is fed back into the correction loop. y ( k Through feedback correction process and y m ( k+j Compare and generate y r ( k + j The output is corrected to correct the prediction model error, and rolling optimization and feedback correction are repeated.

[0024] In the global optimization layer, the grid frequency deviation, frequency change rate, and charge state distribution balance indices are used as the system's state-space equations as follows: ; Where, Δ f t Current time t Frequency deviation; Entropy is the rate of change of frequency. SOC t () is an index of the uniformity of the state of charge distribution; To optimize the state-of-charge distribution balance of energy storage clusters in coordinated frequency regulation, an entropy function is introduced as a quantitative indicator. The aim is to dynamically balance the depth of charge and discharge, thereby synchronizing the capacity decay of the entire system and extending its overall lifespan. The specific definition of the entropy function is: ; ; in, SOC i For the first i The state of charge of each energy storage unit; N This refers to the number of energy storage units; The smaller the entropy function value, the more balanced the distribution of the charged state; The weights of the model predictive control objective function are continuously adjusted based on the system's operating state to balance the relative importance of different objectives. Its action space state expression is as follows: ; in, α t For time t Frequency deviation suppression weights; β t For time t Minimize the weight of energy storage losses; γ t For time t The weights for balancing the state of charge; After executing the model predictive control process, the system provides numerical feedback. The reward function guides deep reinforcement learning (DRL) to learn a multi-objective collaborative policy, and its expression is: ; in, w 1. w 2 and w 3 is the normalization coefficient; P ess For frequency modulation output; The normalization coefficient is used to balance the priorities of various objectives and to penalize frequency deviation, energy storage loss and state of charge imbalance.

[0025] The power coordination layer adopts a model predictive control-based power coordination strategy for wind, solar, and energy storage during black start. Its core principle is to receive model predictive control weight coefficients generated by the upper layer, establish a closed-loop control framework of predictive model, rolling optimization, and feedback correction, and achieve dynamic coordination and optimization of the wind, solar, and energy storage system during black start. The control principle diagram is shown below. Figure 3As shown, based on the predicted values ​​of wind and solar power, the predicted value of photovoltaic power, and the state of charge of energy storage, and based on the power balance equation and the output power change equation of wind farm and photovoltaic power station, the state space equation is established through discretization processing to generate the state prediction sequence of wind-solar-storage microgrid in the future prediction time domain. The prediction time domain consists of P consecutive time steps, and the control time domain consists of M time steps, and satisfies P≥M. Within each control cycle, the optimization problem is solved online based on the predicted sequence. The objective function comprehensively considers the objectives of total wind and solar power output tracking load demand, maintaining the energy storage state of charge within a reasonable range, and minimizing the number of power generation unit switching operations. This yields the optimal control sequence in the control time domain, including the change in the number of photovoltaic units Δ. N pv Change in the number of wind turbines Δ N w and energy storage power P B ; The actual measured output power of the photovoltaic power station P pv Wind farm output power P w The energy storage state of charge output is used as feedback to correct prediction errors and to perform rolling optimization again in the next control cycle to achieve closed-loop control. In actual execution, only the control instruction of the first time step in the control sequence is used. After each step is executed, the system collects the actual output as feedback, updates the state variables, and repeatedly optimizes the process.

[0026] The state-space prediction model of the wind-solar-storage system uses the energy storage charging and discharging power, the output power of the photovoltaic power station, and the output power of the wind farm as state variables, the change in the number of photovoltaic units and wind turbines as control variables, and the black-start load power as an external disturbance. Through discretized state equations, it predicts the future operating state in the time domain. The state-space model equations of the wind-solar-storage microgrid are as follows: ; The state variable is represented as: ; in, P B ( k )for k The charging and discharging power of energy stored at all times; P pv ( k )for k The output power of the photovoltaic power station at any given time; P w ( k )for k The output power of the wind farm at any given time; The control variable is represented as: ; Where, Δ N pv Δ represents the change in the number of photovoltaic cells. N w This represents the change in the number of wind turbines; External disturbance r ( k )= P load , P load For load power; System output y ( k )= P B , P B The charging and discharging power for energy storage; Matrices A, B1, B2, and C are shown below: ; ; ; ; D 1 and D 2 is set as a zero matrix; D 1 represents the instantaneous impact of changes in the number of photovoltaic units and wind turbines on the energy storage power. In actual systems, the number of photovoltaic units and wind turbines needs to gradually affect the energy storage through the power balance equation, rather than having an instantaneous effect. Therefore, the instantaneous impact is non-existent or negligible. D 2. The instantaneous impact of load power fluctuations on energy storage power: Load changes do not physically alter the energy storage output immediately, but rather indirectly affect it through frequency or power deviations, adjusted by the control system. D 2 is set as a zero matrix; When the photovoltaic system is shut down at night, the photovoltaic component in the state-space model equation is set to zero, and the parameter matrix of the wind-storage system can be extracted. Conversely, by shielding the wind component, the corresponding matrix of the photovoltaic-storage system can be obtained. Substituting the two sets of matrices back into the state-space model equation, the prediction models for wind-storage or photovoltaic-storage systems can be constructed.

[0027] In the rolling optimization phase, the system solves a multi-objective optimization problem in each control cycle. The primary optimization objective is to minimize the deviation between the total wind and solar power output and the load power to ensure power balance. The optimization process must satisfy constraints such as power balance constraints, energy storage operation boundary constraints, and the number of power generation units. The specific cost function is shown below: ; Since energy storage output directly affects the system's speed and accuracy in adjusting to frequency deviation, and state-of-charge imbalance can cause differentiated aging of energy storage units, the design objective function includes optimization indices for frequency deviation, energy storage loss, and state-of-charge balance. The specific cost function is shown below: ; At the same time, the switching frequency of power generation units is limited to control the change in the number of photovoltaic units Δ within the control time domain. N PV Change in the number of fans Δ N W Minimize the cost to avoid frequent device start-ups and shutdowns. The cost function is as follows: ; Local optimization of each energy storage unit in distributed energy storage i In the prediction time domain T The system executes the initial control command to perform rolling updates and solves the equation. The equation expression is: ; in, Indicates the first i One energy storage unit from t +1 to t + T Minimize the frequency modulation output sequence within the predicted time domain; t + k | t Indicates time t For future moments t + k The predicted value; Considering the power and state of charge constraints, the following conditions are met: ; ; in ，P rated,i For the first i Rated power of each energy storage unit; For the first i Minimum permissible state of charge for each energy storage unit; For the first i The maximum permissible charge of each energy storage unit; Finally, by using a real-time feedback correction mechanism, the actual operating state of the system is used as the initial condition for the next optimization cycle, effectively compensating for the impact of wind and solar power prediction errors and enhancing the system's robustness in the face of uncertainties. This strategy significantly reduces the depth of energy storage charging and discharging and power fluctuations by coordinating and controlling the number of wind turbines, the number of photovoltaic units and the energy storage power. While ensuring the reliability of black start, it reduces the configuration requirements for energy storage capacity.

Claims

1. A black-start control method for a wind-solar-storage microgrid based on model predictive control, characterized in that, The method includes the following steps: A microgrid structure model for wind, solar and energy storage is established. By adjusting the number of wind turbines in operation, the scale of photovoltaic power station input units, and the discharge power of the energy storage system, the output levels of the wind power generation system and the photovoltaic power generation system are managed. The wind power generation, photovoltaic power generation and energy storage charging and discharging power of the wind-solar-storage microgrid are predicted by a two-layer control architecture, and the energy storage state of charge is adjusted. The two-layer control architecture includes a global optimization layer and a power coordination layer. The global optimization layer adopts a model prediction algorithm based on reinforcement learning, continuously adjusts the weights of the predictive control objective function according to the system operating state, and generates the optimal control model through the reward function; The power coordination layer adopts a model predictive control-based power coordination strategy for black start of wind, solar and energy storage microgrids. It establishes a closed-loop control framework of predictive model, rolling optimization and feedback correction to control the power during the black start process of wind, solar and energy storage microgrids.

2. The black-start control method for wind-solar-storage microgrids based on model predictive control according to claim 1, characterized in that, The specific power output levels of wind power generation systems and photovoltaic power generation systems are as follows: The regulation and control of wind farms and photovoltaic power plants is described as follows: ； in, N pv ( k ) is the target time k The number of photovoltaic units; N pv ( k +1) represents the target time. k +1 to the number of photovoltaic units; Δ N pv ( k (time) k At the time k +1 represents the change in the number of photovoltaic units; N w ( k ) is the target time k The number of wind turbines; N w ( k +1) represents the target time. k +1 number of fans; Δ N w (k) represents time (k) k At the time k +1 represents the change in the number of wind turbines; The process of power output change in wind farms and photovoltaic power plants is as follows: ； in, P pv ( k ) is the target time k Measured output power of photovoltaic power station; P pv ( k +1) represents the target time. k +1 Measured output power of photovoltaic power station; P pvn ( k The output power of the photovoltaic unit is derived from ultra-short-term power prediction. P w ( k ) is the target time k Measured output power of the wind farm; P w ( k +1 represents the target time. k +1 Measured output power of wind farm; P wn ( k The output power of the wind turbine is derived from the ultra-short-term power prediction.

3. The black-start control method for a wind-solar-storage microgrid based on model predictive control according to claim 1, characterized in that, The conversion process of electricity in an energy storage system can be represented as follows: ； in, E B ( k ) is the target time k Real-time energy storage capacity; E ( k +1) represents the target time. k +1 real-time energy storage capacity; η The conversion efficiency of energy storage; Δ T B This is the conversion factor from power to electrical quantity; P B ( k ) is the target time k Energy storage capacity.

4. The black-start control method for wind-solar-storage microgrids based on model predictive control according to claim 1, characterized in that, The reinforcement learning-based model prediction algorithm is specifically as follows: A reference trajectory is generated based on the current state of the wind-solar-storage microgrid system. y s ( k Based on machine learning models, future dynamics are predicted to obtain energy storage forecast values. y m ( k+j ); Optimize the first step control amount of the preset scrolling. u ( k The input is the rolling optimization module, which solves for the optimal control sequence based on the objective function, applies it to the controlled object, and outputs the actual value of energy storage. y ( k ); Sensors collect actual energy storage values y ( k The input is fed into the neural network to calculate the state-action value function Q(s|a), and an evaluation signal is given based on the reward function of the rolling optimization module. The control parameters or strategies of the model predictive control are dynamically adjusted, and the control decision is fed back into the correction loop. y ( k Through feedback correction process and y m ( k+j Compare and generate y r ( k + j The output is corrected to correct the prediction model error, and rolling optimization and feedback correction are repeated.

5. A black-start control method for a wind-solar-storage microgrid based on model predictive control according to claim 2 or 3, characterized in that, In the global optimization layer, the grid frequency deviation, frequency change rate, and charge state distribution balance index are used as the system's state-space equations, as follows: ； Where, Δ f t Current time t Frequency deviation; Entropy is the rate of change of frequency. SOC t () is an index of the uniformity of the state of charge distribution; The weights of the model predictive control objective function are continuously adjusted based on the system's operating state to balance the relative importance of different objectives. Its action space state expression is as follows: ； in, α t For time t Frequency deviation suppression weights; β t For time t Minimize the weight of energy storage losses; γ t For time t The weights for balancing the state of charge; After executing the model predictive control flow, the system provides numerical feedback. The reward function guides the DRL to learn a multi-objective cooperative strategy, and its expression is: ； in, w 1. w 2 and w 3 is the normalization coefficient; P ess For frequency modulation output; The normalization coefficient is used to balance the priorities of various objectives and to penalize frequency deviation, energy storage loss and state of charge imbalance.

6. A black-start control method for a wind-solar-storage microgrid based on model predictive control according to claim 2 or 3, characterized in that, The state-space model equation of the power coordination layer prediction model is: ； The state variable is represented as: ； in, P B ( k )for k The charging and discharging power of energy stored at all times; The control variable is represented as: ； External disturbance quantity r ( k )= P load , P load For load power; System output y ( k )= P B , P B The charging and discharging power for energy storage; Matrices A, B1, B2, and C are shown below: ； ； ； ； D 1 and D 2 is set as a zero matrix.

7. A black-start control method for a wind-solar-storage microgrid based on model predictive control according to claim 6, characterized in that, The prediction model for a wind power generation system or a photovoltaic power generation system can be obtained by shielding another subsystem and setting its matrix components to zero, and then substituting them into the state-space model equation.

8. A black-start control method for a wind-solar-storage microgrid based on model predictive control according to claim 5, characterized in that, The cost function for the rolling optimization is as follows: ； Where M is the time step; P load For load power; The cost function, which includes optimization metrics such as frequency deviation, energy storage loss, and state-of-charge balance, is shown below: ； The cost function for limiting the frequency of switching is shown below: ； Local optimization of each energy storage unit in distributed energy storage i In the prediction time domain T The system executes the initial control command to perform rolling updates and solves the equation. The equation expression is: ； in, Indicates the first i One energy storage unit from t +1 to t + T Minimize the frequency modulation output sequence within the predicted time domain; t + k | t Indicates time t For future moments t + k The predicted value; Considering the power and state of charge constraints, the following conditions are met: ； ； in ，P rated,i For the first i Rated power of each energy storage unit; For the first i Minimum permissible state of charge for each energy storage unit; For the first i The maximum permissible charge of each energy storage unit.

9. A black-start control method for a wind-solar-storage microgrid based on model predictive control according to claim 5, characterized in that, The entropy function is introduced to quantify the uniformity index of the state of charge distribution. The specific definition of the entropy function is: ； ； in, SOC i For the first i The state of charge of each energy storage unit; N This represents the number of energy storage units.

10. A black-start control method for a wind-solar-storage microgrid based on model predictive control according to claim 8, characterized in that, By using a real-time feedback correction mechanism, the actual operating status of the system is used as the initial condition for the next optimization cycle to compensate for the impact of wind and solar power prediction errors.

Images