Intelligent scheduling control method for optical storage charging and discharging integrated power station

Abstract

The invention provides an intelligent scheduling control method for an optical storage charging and discharging integrated power station, and the method constructs a multi-time-scale prediction-optimization-correction three-stage control architecture, and achieves the real-time scheduling of the optical storage charging and discharging integrated power station through the coordinated operation in the upper-layer 24-hour rolling optimization, the middle-layer real-time scheduling and the lower-layer equipment control. And multi-target and multi-time-scale coordinated control optimization of maximization of the photovoltaic consumption rate, minimization of the power grid electricity purchase cost and prolonging of the service life of the energy storage system is realized. And a model prediction control algorithm is combined to dynamically adjust photovoltaic power generation output, an energy storage charging and discharging plan and a charging pile scheduling strategy. In addition, an adaptive parameter adjustment mechanism is introduced, and MPC parameters are adjusted online by using a reinforcement learning algorithm so as to cope with dynamic changes of environmental parameters and equipment states. The method can effectively improve the operation efficiency and economy of the optical storage charging and discharging integrated power station, prolongs the service life of an energy storage system, and is suitable for intelligent scheduling control of various optical storage charging and discharging integrated power stations.

Description

Technical Field

[0001] The present invention relates to the field of intelligent dispatching control of power systems, and particularly to an intelligent dispatching control method for a photovoltaic-storage-charging integrated power station. Background Art

[0002] Under the background of the global energy structure transformation towards low-carbon and intelligent, the photovoltaic-storage-charging integrated power station has gradually become an important part of the smart energy system. Its core goal is to achieve efficient utilization of photovoltaic power generation, and through the collaborative optimization of the energy storage system and charging infrastructure, improve the overall operation efficiency of the power station. With the rapid development of renewable energy and the popularization of electric vehicles, the photovoltaic-storage-charging integrated power station, as a new type of integrated energy system, is of great significance in improving energy utilization efficiency and reducing carbon emissions.

[0003] However, due to the intermittency of photovoltaic power generation, the randomness of electric vehicle charging load, and the volatility of electricity prices, the operation of the photovoltaic-storage-charging integrated power station faces many challenges. The traditional photovoltaic-storage-charging management method mainly relies on rule-based control strategies, which fail to fully consider the randomness of photovoltaic power generation, the dynamic charging demand of electric vehicles, and the health management of the energy storage system, resulting in low photovoltaic power consumption rate, serious light abandonment phenomenon, poor operation economy, failure to effectively utilize electricity price fluctuations to optimize the power purchase plan, rapid energy storage life attenuation, and lack of dynamic adjustment ability of control strategies, making it difficult to cope with load and environmental changes. Photovoltaic power generation is affected by factors such as weather and day-night changes, with strong volatility and high uncertainty, resulting in low photovoltaic power consumption rate and affecting the overall economic benefits of the power station. At the same time, the traditional energy storage system lacks intelligent management, and the charge-discharge strategy does not fully consider the battery life degradation characteristics, accelerating battery attenuation and reducing long-term economy. Under the peak-valley electricity price mechanism, the peak-shaving and valley-filling characteristics of the energy storage system are not fully utilized, and the optimal power purchase strategy cannot be achieved, increasing the power grid power purchase cost. The charging demand of electric vehicles has great randomness. Without a reasonable dispatching strategy, it may cause local load overload and affect the safe and stable operation of the power station. The control strategies of existing photovoltaic-storage-charging power stations generally have problems such as low photovoltaic power consumption rate, poor economy, rapid energy storage life attenuation, and insufficient adaptability of control parameters, manifested as difficulty in effectively coping with photovoltaic power generation fluctuations, ignoring electricity price fluctuations, frequent energy storage charge-discharge accelerating battery attenuation, and inability to dynamically adjust control parameters according to real-time status and environmental changes.

[0004] At present, traditional control methods often struggle to simultaneously take into account the fluctuations of photovoltaic power generation, the dynamic characteristics of energy storage systems, and the real-time requirements of charging pile scheduling. Moreover, they cannot achieve cost optimization under the background of electricity price fluctuations. There is a lack of an intelligent control strategy in the existing technologies that can comprehensively consider the real-time load of the power station, meteorological parameters, and electricity price fluctuations, and achieve multi-objective collaborative optimization through predictive control algorithms. Therefore, there is an urgent need for a new type of intelligent control strategy that can not only improve the photovoltaic power consumption rate, but also reduce the grid power purchase cost, extend the service life of the energy storage system, ensure the safety and reliability of system operation, and have good adaptability, so as to significantly improve the operation efficiency and economy of the integrated photovoltaic energy storage charging and discharging power station.

[0005] After retrieval, the invention patent with the Chinese patent publication number CN118693817B proposes an energy scheduling system and method for an integrated photovoltaic energy storage charging and discharging power station based on robust control. In this invention, an AC bus coupling system is used as the framework solution, that is, the in-station power grid of the charging station in the integrated photovoltaic energy storage charging and discharging power station of this system uses an AC bus, and the energy scheduling of the integrated photovoltaic energy storage charging and discharging power station is carried out through a method based on robust control. To cope with the uncertainties such as photovoltaic power output and load demand in the planning of the integrated photovoltaic energy storage charging and discharging power station, a model based on the robust control method of energy change is established, which can realize the orderly coordination of the "source-grid-load-storage" elements on the basis of the uncertainties of photovoltaic power generation and electricity load; it can ensure the stability and economy of the operation of the integrated photovoltaic energy storage charging station and improve the comprehensive income.

[0006] The technical comparison between the above-mentioned comparative document and the present application is as follows:

[0007] The intelligent scheduling control strategy design method for the integrated photovoltaic energy storage charging and discharging power station proposed in this patent constructs a three-level coordinated control architecture of prediction-optimization-correction at multiple time scales, forming an organic connection between the upper-layer 24-hour rolling optimization, the middle-layer real-time scheduling, and the lower-layer device control, thus realizing the multi-objective coordinated optimization of maximizing the photovoltaic power consumption rate, minimizing the grid power purchase cost, and extending the service life of the energy storage system; while the patent CN118693817B mainly adopts an AC bus coupling scheme based on robust control, focusing on realizing the orderly coordination of each link of "source-grid-load-storage" under the uncertain conditions of photovoltaic power output and load demand.

[0008] Second, in the design of the scheduling strategy of the present invention, model predictive control and reinforcement learning algorithms are introduced, and the online adaptive parameter adjustment mechanism is used to dynamically respond to changes in the environment and device states, so as to more accurately adjust the photovoltaic power output, the energy storage charging and discharging plan, and the charging pile scheduling strategy, further improving the system operation efficiency and economy; in contrast, the patent CN118693817B focuses on using robust control to ensure the stability and economic benefits of system operation, and there are obvious differences in the technical focus and implementation means.

[0009] After retrieval, the invention patent with the Chinese patent publication number CN111628493A proposes an integrated energy storage power station control method for photovoltaic energy storage, and its control method is a comprehensive energy automatic control method that integrates photovoltaic power generation, energy storage power station, and load power consumption. It includes an automatic control method for integrating the photovoltaic energy storage power station with the power grid, an automatic control method for integrating the photovoltaic energy storage power station with industry and commerce, and an automatic control method for integrating the photovoltaic energy storage power station with new energy. The present invention relates to the field of electric power technology. This integrated energy storage power station control method realizes power grid peak shaving and frequency modulation, participates in the peak shaving and frequency modulation of the power grid, promotes the gradual withdrawal of coal-fired peak shaving units, and the integrated photovoltaic energy storage power station obtains benefits through peak shaving compensation, realizes demand-side response, and ensures the stability of the power grid when the power grid power supply system avoids peak electricity consumption or interrupts power. Therefore, only the energy storage device is an effective means to achieve demand response and also an effective way to solve the contradiction between power supply and demand.

[0010] The technical comparison between the above-mentioned comparative document and the present application is as follows:

[0011] 1. This patent realizes the coordinated optimization of multiple system objectives through a multi-level and cross-time-scale intelligent scheduling control strategy, comprehensively considering photovoltaic power generation, energy storage charging and discharging, and charging pile scheduling; while patent CN111628493A adopts a comprehensive energy automatic control method, integrating photovoltaic, energy storage, and load power consumption, mainly focusing on power grid peak shaving and frequency modulation and demand response to obtain economic benefits.

[0012] 2. The present invention emphasizes that in a dynamic environment, real-time adjustment of parameters is achieved through model predictive control combined with online reinforcement learning, so as to flexibly respond to the fluctuations of photovoltaic output and load demand, and ensure the long-term healthy operation of the energy storage system; in contrast, patent CN111628493A mainly relies on a fixed control logic to achieve system adjustment, and it is difficult to make timely responses to the rapid changes in the environment and equipment status.

[0013] After retrieval, the invention patent with the Chinese patent publication number CN118693817A proposes an energy scheduling system and method for an integrated photovoltaic energy storage charging station based on robust control. In the present invention, an AC bus coupling system is used as the framework solution, that is, in the integrated photovoltaic energy storage charging station of this system, the internal power grid of the charging station uses an AC bus, and the energy scheduling of the integrated photovoltaic energy storage charging station is carried out through a method based on robust control. To cope with the uncertainties such as photovoltaic output and load demand in the planning of the integrated photovoltaic energy storage charging station, a model based on a robust control method for energy change is established, which can realize the orderly coordination of the "source-grid-load-storage" elements on the basis of the uncertainties of photovoltaic power generation and power consumption load; it can ensure the stability and economy of the operation of the integrated photovoltaic energy storage charging station and improve the comprehensive benefits.

[0014] The technical comparison between the above-mentioned comparative document and the present application is as follows:

[0015] 1. This patent adopts a three - level coordinated control architecture and a multi - time - scale optimization method, which not only focuses on short - term real - time scheduling but also takes into account 24 - hour rolling optimization, comprehensively improving the comprehensive benefits of photovoltaic accommodation, power grid purchase cost control, and energy storage life extension; while patent CN118693817A realizes the coordinated operation of "source - grid - load - storage" elements under uncertain conditions through AC bus coupling and a robust control model based on energy change, focusing on ensuring the stability and economy of the system.

[0016] 2. While realizing intelligent scheduling, the present invention innovatively introduces a reinforcement learning algorithm for online parameter adjustment, enabling the control strategy to adapt to the dynamically changing external environment and internal state, with higher flexibility and optimization space; in contrast, patent CN118693817A relies on a pre - designed robust control model, and its adaptability and refined control ability in dealing with complex dynamic changes are relatively limited. Summary of the Invention

[0017] To solve the above - mentioned technical problems, the present invention proposes an intelligent scheduling control method for a photovoltaic - energy - storage - charging - discharging integrated power station. By introducing a three - level control architecture with multiple time scales, a model predictive control algorithm, and an adaptive parameter adjustment mechanism, the power station can achieve highly intelligent operation management, fully exert the operation advantages of the photovoltaic - energy - storage - charging - discharging integrated power station, maximize the photovoltaic accommodation rate, minimize the power grid purchase cost, and extend the service life of the energy storage system, thereby improving the overall economy and safety of the power station.

[0018] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0019] An intelligent scheduling control method for a photovoltaic - energy - storage - charging - discharging integrated power station, characterized by comprising the following steps:

[0020] Step S1: Construct a three - level control framework:

[0021] Set three levels: the upper - layer rolling optimization layer, the middle - layer real - time scheduling layer, and the lower - layer device control layer.

[0022] Upper layer: Rolling optimization layer

[0023] The rolling optimization layer is responsible for performing rolling optimization within a time range. With a time resolution of 15 minutes, according to the predicted photovoltaic power generation output, load demand, energy storage charging and discharging requirements, and electricity price fluctuations, it calculates the optimal control strategy within the next 24 hours.

[0024] Considering multiple optimization objectives, design a multi - objective optimization function, and the formula is as follows:

[0025]

[0026] Wherein:

[0027] P PV (t) is the output of photovoltaic power generation;

[0028] P load (t) is the power demand of the power station;

[0029] C(t) is the electricity price at time t;

[0030] SOC deviaton (t) is the deviation between the energy storage SOC and the target SOC;

[0031] P storage is the real-time charge and discharge power of the energy storage device;

[0032] λ1, λ2, λ3, λ4 are the weight coefficients in the objective function, which are adjusted according to the actual application situation,

[0033] By optimizing the objective function, the energy storage charge and discharge strategy and the photovoltaic power generation scheduling strategy within the next 24 hours can be obtained;

[0034] Middle layer: Real-time scheduling layer

[0035] Based on the optimization results of the upper layer, the real-time scheduling layer adjusts the energy storage and photovoltaic power generation output under real-time dynamic conditions through a mathematical model. The optimization objective is to dynamically adjust the charge and discharge plan of the energy storage device,

[0036] Mathematical model: The following real-time scheduling optimization model is adopted, and the formula is as follows:

[0037]

[0038] Wherein:

[0039] P EV (t) is the charging power of the electric vehicle at the current moment;

[0040] P EV,demand (t) is the charging demand of the electric vehicle;

[0041] P storage,optimal (t) is the optimal energy storage charge and discharge power calculated by the upper layer optimization;

[0042] Lower layer: Equipment control layer

[0043] The equipment control layer realizes the specific equipment execution through the control model, and precisely controls the energy storage device, the photovoltaic power generation system and the grid interface in real time,

[0044] The control model is as follows:

[0045] Pstorage,command P(t) = P storage (t) + ΔP storage (t)

[0046] Where:

[0047] P storage,command (t) is the control command of the energy storage device;

[0048] ΔP storage (t) is the fine-tuning command provided by the middle-layer control,

[0049] The constraint conditions of the system are as follows

[0050] Grid power purchase power constraint:

[0051] P grid,min ≤ P grid (t) ≤ P grid,max

[0052] Photovoltaic power generation power constraint:

[0053] 0 ≤ P PV (t) ≤ P PV,max (t)

[0054] Energy storage charge and discharge power constraint:

[0055] 0 ≤ P storage (t) ≤ P storage,max

[0056] Energy storage SOC constraint:

[0057] SOC min ≤ SOC(t) ≤ SOC max ;

[0058] Step S2, establish a prediction model

[0059] Use a long short-term memory network model to predict the photovoltaic power output, load demand and electricity price at future times;

[0060] Step S3, through the prediction model control algorithm for rolling optimization, dynamically adjust the control strategy to minimize the multi-objective optimization function, specifically as follows:

[0061] In the upper-layer optimization, comprehensively consider the photovoltaic accommodation rate, grid power purchase cost and energy storage life, and calculate the optimal control plan within the next 24 hours;

[0062] The middle-layer control then makes fine adjustments at the minute level according to the real-time scheduling requirements to ensure that the system operation is consistent with the real-time optimization goal;

[0063] The lower - layer control is responsible for the specific execution of the equipment. It adjusts the output of the energy storage device and the photovoltaic system through a one - second - level response to meet the optimization strategy in real time;

[0064] Step S4: Establish a state - space model to describe the dynamic response of the power station system,

[0065] Specifically including:

[0066] x(t): State variable,

[0067] u(t): Control variable,

[0068] y(t): Output variable,

[0069] The form of the state - space model is:

[0070] x(t + 1) = Ax(t)+Bu(t)+Ew(t)

[0071] y(t) = Cx(t)+Du(t)

[0072] Where:

[0073] A, B, C, D are system parameter matrices;

[0074] w(t) is an external disturbance, including environmental parameters and load fluctuations;

[0075] Step S5: Adaptive parameter adjustment mechanism

[0076] An enhanced learning algorithm is used to online - adjust the control parameters. The enhanced learning algorithm, through a trial - and - feedback mechanism, autonomously adjusts the weight coefficients of the prediction model and dynamically optimizes the control strategy under different working conditions, specifically as follows:

[0077] S51: Define the state space. Take the operation state information of the power station as the input state of the enhanced learning algorithm. The state variables include:

[0078] s t =[P PV (t), SOC(t), P EV (t), T out (t), C(t)]

[0079] Where

[0080] T out (t): Environmental temperature at time t;

[0081] S52: Define the action space. Take the control parameters of the model predictive control algorithm as the output action of the enhanced learning algorithm. The defined control parameters include:

[0082] λ t= [λ1(t), λ2(t), λ3(t), λ4(t), N(t)]

[0083] where

[0084] N(t): the rolling optimization time domain of the model predictive control algorithm, the control prediction time range;

[0085] S53: Define the reward function. According to the optimization goal of the integrated photovoltaic energy storage charging and discharging power station, define the reward function r t to guide the reinforcement learning algorithm to optimize the control parameters

[0086] r t = ω1·η PV (t) - ω2·P grid (t) - ω3·DOD(t)

[0087] where:

[0088] ω1, ω2, ω3 are the weight coefficients of the reward function

[0089] η PV (t) represents the photovoltaic accommodation rate

[0090] D battery (t) is the charge and discharge depth of the energy storage battery, indicating the discharge degree of the battery;

[0091] S54: Determine the goal of reinforcement learning, that is, find the optimal policy π * , select the optimal action at each moment t to maximize the cumulative reward:

[0092]

[0093] where:

[0094] π represents the policy function, defining the mapping between the state space and the action space;

[0095] γ is the discount factor, used to balance short-term rewards and long-term benefits;

[0096] r t is the immediate reward at each moment;

[0097] S55: Adopt the experience replay mechanism to break the time correlation through random sampling and eliminate the correlation between data;

[0098] S56: Perform control parameter update. In each control cycle, the reinforcement learning algorithm updates the control parameters of the model predictive control algorithm according to the operating state and prediction data of the power station. If the reward value increases, the control parameters remain unchanged; if the reward value decreases, the control parameter combination is adjusted to find a better solution;

[0099] Step S6, Real-time Feedback and Adjustment

[0100] Monitor the operating status of power station equipment and environmental parameters in real time. The results of real-time monitoring will be combined with the output of the model predictive control algorithm to ensure that the system can dynamically adjust the control strategy according to the changes in power station equipment parameters. When the power station equipment or environmental parameters change, the control system can quickly adjust the charge and discharge strategy of the energy storage system, the scheduling plan of photovoltaic power generation, and the decision-making of power grid power purchase according to the real-time feedback information, so as to optimize the overall operating status of the system in the shortest time.

[0101] As a preferred technical solution of the present invention: in step S1, the time resolution of the rolling optimization layer is 15 minutes, the time resolution of the real-time scheduling layer is 1 minute, and the time resolution of the equipment control layer is 1 second.

[0102] As a preferred technical solution of the present invention: the rolling optimization layer is a 24-hour rolling optimization.

[0103] As a preferred technical solution of the present invention: in step S2, the prediction model includes a photovoltaic output prediction model, an energy storage dynamic model, and a charging demand response model to achieve photovoltaic output prediction, energy storage charge and discharge demand response, and electric vehicle charging demand response.

[0104] Photovoltaic Output Prediction Model:

[0105] Use a machine learning model based on historical data to predict the photovoltaic power generation. By training and optimizing the input of the photovoltaic output prediction model, an accurate future photovoltaic output prediction can be obtained. The formula is as follows:

[0106] P PV,pred (t) = f weather (t) · P PV,max

[0107] Where:

[0108] P PV,pred (t) is the predicted photovoltaic power generation

[0109] f weather (t) is the light function calculated according to meteorological data

[0110] P PV,max is the maximum output power of the photovoltaic system;

[0111] Energy Storage Dynamic Model:

[0112] Establish an energy storage dynamic model of the energy storage system, and obtain the real-time operating status of the energy storage system through dynamic simulation. The formula is as follows:

[0113]

[0114] Wherein:

[0115] E storage (t + 1) is the energy storage power at the next moment

[0116] P charge (t), P discharge (t) are the charging and discharging powers

[0117] η charge and η discharge are the charging and discharging efficiencies respectively;

[0118] Charging demand response model:

[0119] Based on the scheduling requirements of electric vehicle charging piles, a charging demand response algorithm is designed. According to different load demands and electricity price fluctuations, the use of charging piles is reasonably scheduled. The formula is as follows:

[0120] P EV,demand (t) = f EV (t) · P EV,max

[0121] Wherein:

[0122] P EV,demand (t) is the charging demand of the electric vehicle

[0123] f EV (t) is the demand adjustment factor based on real-time load prediction

[0124] P EV,max is the maximum power for electric vehicle charging.

[0125] As a preferred technical solution of the present invention: In step S55, the experience replay mechanism is specifically as follows:

[0126] S551: Initialize the experience pool, and create an experience pool D with a capacity of D max ;

[0127] S552: In each control cycle, record the state, action, reward, and next state at the current moment to form a quadruple sample (s t , a t , r t , s t+1 ), and store this sample in the experience pool D; if the experience pool is full, dequeue the earliest sample;

[0128] S553: During the model training process, randomly extract a sample from the experience pool, and set the number of samples to N; update the value network according to the following objective function

[0129]

[0130] Among them, L(θ) is the loss function;

[0131] Q′ and π′ are the Q-value function and policy function of the target network respectively.

[0132] S454: Update the Actor network parameters using policy gradients according to the sampled samples:

[0133]

[0134] As a preferred technical solution of the present invention: In step S552, the PER mechanism is introduced to assign priorities to the samples in the experience pool, and samples that contribute more to policy improvement are preferentially selected, thereby accelerating the convergence speed of the model. The priority calculation formula of PER is as follows

[0135]

[0136] Among them,

[0137] δ i is the error of sample i

[0138] α is a hyperparameter that controls the priority assignment weight of the samples.

[0139] As a preferred technical solution of the present invention: In step S56, the control parameters are updated in real time through the following dynamic adjustment formula:

[0140]

[0141] Among them:

[0142] θ opt (t) is the optimal control parameter at the current moment

[0143] θ prev (t) is the control parameter at the previous moment

[0144] η is the learning rate, which is used to control the update step size

[0145] is the gradient of the control objective function with respect to the control parameter, indicating the sensitivity of the control parameter to the optimization objective.

[0146] Compared with the prior art, the beneficial effects of the present invention are:

[0147] The intelligent control method of the present invention has significant advantages. By combining model predictive control and reinforcement learning algorithms, it can achieve precise scheduling and dynamic optimization of the integrated photovoltaic energy storage charging and discharging power station. This method can automatically adjust various control parameters on the premise of meeting multiple objectives such as photovoltaic power consumption, grid power purchase cost, and energy storage battery life, ensuring that the system operates in the optimal state, thereby significantly improving the operation efficiency and economy of the power station.

[0148] The prediction module based on the LSTM model significantly improves the prediction accuracy of time series data such as photovoltaic power output and load fluctuations, providing an accurate basis for the optimization of control strategies. In terms of multi-objective comprehensive optimization, the MPC algorithm is used to ensure the balance of photovoltaic power consumption, grid power purchase cost, and battery life, thereby realizing the optimal control strategy and improving the overall system efficiency.

[0149] In addition, the introduction of the reinforcement learning algorithm realizes the adaptive adjustment of control parameters, enabling the system to quickly respond to changes in the power station operation state and environmental parameters, improving the stability and response speed of the system. Finally, the control strategy has good modularity, parameter self-tuning, and fast response characteristics, and is applicable to various integrated photovoltaic energy storage charging and discharging power stations. It not only improves the flexibility of the system but also can be widely promoted and applied in different types of integrated photovoltaic energy storage charging and discharging power stations, showing strong adaptability and scalability. The present invention introduces a real-time monitoring and feedback mechanism. By collecting and analyzing the power station equipment status and environmental parameters in real time, it can dynamically adjust the control strategy to cope with emergencies such as load fluctuations and weather changes, improving the power station's adaptability and response speed to external environmental changes. This can effectively avoid equipment overload or energy waste, further improving the safety and reliability of the power station.

[0150] In addition, the experience replay mechanism and prioritized experience replay mechanism adopted by the present invention, by making full use of historical data and experience samples, can not only accelerate the learning process, improve the sample utilization rate, but also enhance the generalization ability of the model, enabling it to better adapt to the complex and changeable power station operation environment, thereby improving the stability and long-term optimization ability of the control strategy. In terms of equipment management, the intelligent control system of the present invention maximizes the utilization rate of power station resources by precisely scheduling equipment such as photovoltaic power generation, energy storage systems, and electric vehicle charging piles. Its modular hardware design and flexible control strategy can adapt to power station applications of different scales, with strong scalability and compatibility, providing a guarantee for the sustainable development of the long-term operation and management of the power station.

[0151] The present invention not only improves the economy and operation efficiency of the integrated photovoltaic energy storage charging and discharging power station, but also shows significant advantages in terms of system stability, safety, and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0152] Figure 1It is the design diagram of the intelligent scheduling control strategy architecture in the present invention;

[0153] Figure 2 It is a schematic diagram of the structure of the long short-term memory network of the prediction model in the present invention. DETAILED DESCRIPTION

[0154] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0155] like Figure 1-2 As shown, the present invention proposes an intelligent dispatching and control method for an integrated photovoltaic storage and charging and discharging power station, which adopts a three-level control architecture of "prediction-optimization-correction" and ensures that the power station can be accurately optimized and controlled on different time scales through coordination and division of labor at the upper, middle and lower levels.

[0156] The upper layer is the rolling optimization layer, the middle layer is the real-time scheduling layer, and the lower layer is the device control layer.

[0157] Upper layer: scroll optimization layer

[0158] The upper control is responsible for long-term (24-hour) rolling optimization, with a time resolution of 15 minutes, and calculates the optimal control strategy for the next 24 hours based on the predicted photovoltaic power generation output, load demand, energy storage charging and discharging requirements, and electricity price fluctuations. The optimization goals include maximizing the photovoltaic absorption rate, using photovoltaic power generation to provide electricity as much as possible, and reducing the need to purchase electricity from the power grid; minimizing the cost of purchasing electricity from the power grid, reducing the cost of purchasing electricity by reasonably scheduling the energy exchange between the energy storage system and the power grid; and extending the service life of the energy storage system by controlling the depth of charge and discharge to avoid overcharging or over-discharging of energy storage equipment.

[0159] Taking into account multiple optimization objectives at the same time, a multi-objective optimization function is designed, and the formula is as follows:

[0160]

[0161] in:

[0162] P PV (t) is the photovoltaic power output;

[0163] P load (t) is the load demand of the power station;

[0164] C(t) is the electricity price at time t;

[0165] SOC deviaton (t) is the deviation between the energy storage SOC and the target SOC;

[0166] P storage The real-time charging and discharging power of the energy storage device;

[0167] λ1, λ2, λ3, and λ4 are the weight coefficients in the objective function and are adjusted according to the actual application scenario.

[0168] By optimizing this objective function, the energy storage charge-discharge strategy and photovoltaic power generation scheduling strategy within the next 24 hours can be obtained.

[0169] Middle layer: Real-time scheduling layer

[0170] The middle layer control has a time resolution of 1 minute. Based on the optimization results of the upper layer, it adjusts the energy storage and photovoltaic power generation output under real-time dynamic conditions to ensure that the operation of the power station conforms to the current load demand and grid conditions. The optimization objectives are to dynamically adjust the charge-discharge plan of the energy storage device, make the best use of photovoltaic power generation, and reduce grid power purchase; ensure that the charging demand of electric vehicles (EVs) is met without causing grid overload.

[0171] Mathematical model: In this layer, the following real-time scheduling optimization model is adopted, and the formula is as follows:

[0172]

[0173] Where:

[0174] P EV (t) is the charging power of the electric vehicle at the current moment;

[0175] P EV,demand (t) is the charging demand of the electric vehicle;

[0176] P storage,optimal (t) is the optimal energy storage charge-discharge power calculated from the upper layer optimization.

[0177] This layer ensures the minimization of the grid power purchase cost and meets the electric vehicle charging demand by adjusting the output power of the photovoltaic and energy storage systems in real time.

[0178] Lower layer: Equipment control layer

[0179] The lower layer control is responsible for the specific equipment execution with a time resolution of 1 second, and precisely controls the energy storage device, photovoltaic power generation system, and grid interface in real time. The objectives are to accurately control the charge-discharge operation of the energy storage at the 1-second level to avoid over-discharge or over-charging; adjust the photovoltaic power generation output to ensure the load balance of the power station. The objective is that the lower layer control performs precise operations according to the real-time status of the equipment.

[0180] The control model is as follows:

[0181] P storage,command (t) = P storage (t) + ΔP storage (t)

[0182] Where:

[0183] P storage,command P(t) is the control command for the energy storage device;

[0184] ΔP storage ΔP(t) is the fine-tuning command provided by the middle-layer control.

[0185] This control system can respond quickly within milliseconds, ensuring the maximization of the charge and discharge efficiency of the energy storage system and guaranteeing the stability of the system.

[0186] The constraint conditions of the system are as follows

[0187] Grid power purchase power constraint:

[0188] P grid,min ≤P grid (t)≤P grid,max

[0189] Photovoltaic power generation power constraint:

[0190] 0≤P PV (t)≤P PV,max (t)

[0191] Energy storage charge and discharge power constraint:

[0192] 0≤P storage (t)≤P storage,max

[0193] Energy storage SOC constraint:

[0194] SOC min ≤SOC(t)≤SOC max

[0195] In order to comprehensively consider multiple objectives, a multi-objective optimization function is designed. This function balances the priorities among various objectives through the adjustment of weight coefficients, thereby achieving the optimal allocation of power station resources.

[0196] Through the model predictive control algorithm, combined with photovoltaic output prediction, energy storage charge and discharge demand, and electric vehicle charging demand, the charge and discharge strategy of the power station is adjusted in real time. In the middle-layer and lower-layer control systems, the photovoltaic power generation output and the charge and discharge strategy of the energy storage device are adjusted in real time to ensure the minimization of the grid power purchase cost and the maximization of the service life of the energy storage system.

[0197] The objective function of PV accommodation rate is used to measure the utilization efficiency of photovoltaic power generation and maximize the photovoltaic power generation as much as possible. The meaning of this objective function is to maximize the satisfaction of load demand by photovoltaic power generation and reduce the demand for power purchase from the grid. The objective function of grid power purchase cost is used to minimize the cost of grid power purchase. By controlling the scheduling strategies of energy storage and photovoltaic power generation, the power purchase from the grid is minimized as much as possible. By minimizing this objective function, the total cost of power purchase from the grid by the power station can be reduced, especially during periods of high electricity prices, and energy storage and photovoltaic power generation are preferentially used. The objective function of energy storage life is used to extend the service life of the energy storage system, avoid overcharging or over-discharging, and thus reduce battery degradation. The energy storage life is closely related to the charge-discharge depth of the battery. The goal is to keep the battery in the best working state by reasonably scheduling charge-discharge operations. The role of this objective function is to avoid overcharging and over-discharging by controlling the charge-discharge depth of the energy storage system and extend the service life of the energy storage battery.

[0198] During the optimization process, the MPC (Model Predictive Control) algorithm dynamically adjusts the control strategy through rolling optimization technology to minimize the multi-objective optimization function.

[0199] The specific optimization process is as follows:

[0200] In the upper-layer optimization, considering the PV accommodation rate, grid power purchase cost, and energy storage life comprehensively, the optimal control scheme within the next 24 hours is calculated. The middle-layer control then makes fine adjustments at the minute level according to real-time scheduling requirements to ensure that the system operation is consistent with the real-time optimization goal. The lower-layer control is responsible for the specific execution of the equipment, and adjusts the output of the energy storage equipment and the photovoltaic system through 1-second-level response to meet the optimization strategy in real time.

[0201] By designing such a multi-objective optimization function, the present invention can consider multiple factors simultaneously, such as PV accommodation, grid power purchase cost, and energy storage life, ensure the extension of the service life of the energy storage equipment while maximizing PV utilization and minimizing grid power purchase, and achieve the efficient, economic, and sustainable operation of the power station.

[0202] The prediction model uses a Long Short-Term Memory (LSTM) model to predict the photovoltaic power generation output, load demand, and electricity price at future moments; the LSTM model significantly improves the prediction accuracy of photovoltaic fluctuations and load changes by memorizing time series features.

[0203] In step S2, the prediction model includes a photovoltaic output prediction model, an energy storage dynamic model, and a charging demand response model to achieve photovoltaic output prediction, energy storage charge-discharge demand response, and electric vehicle charging demand response.

[0204] Among them: Photovoltaic output prediction model: A machine learning model based on historical data is used to predict the photovoltaic power generation. The model inputs include meteorological data, historical light intensity, etc. Through training and optimization, accurate future photovoltaic output predictions are obtained.

[0205] P PV,pred (t) = f weather (t)·P PV,max

[0206] Among them:

[0207] P PV,pred (t) is the predicted photovoltaic power generation;

[0208] f weather (t) is the light function calculated based on meteorological data;

[0209] P PV,max is the maximum output power of the photovoltaic system.

[0210] Through the combination of historical data and meteorological data, this model can predict the photovoltaic power generation capacity in the future for a period of time.

[0211] Energy storage dynamic model: An energy storage system charge and discharge model is established, considering parameters such as the charge and discharge efficiency, cycle life, and remaining capacity of the energy storage device. Through dynamic simulation, the real-time operating state of the energy storage system is obtained.

[0212]

[0213] Among them:

[0214] E storage (t + 1) is the energy storage power at the next moment;

[0215] P charge (t), P discharge (t) are the charge and discharge powers;

[0216] η charge 、η discharge are the charge and discharge efficiencies respectively.

[0217] Charging demand response model: Based on the scheduling requirements of electric vehicle charging piles, a charging demand response algorithm is designed. According to different load demands and electricity price fluctuations, the use of charging piles is reasonably scheduled to avoid excessive load during peak hours.

[0218] P EV,demand (t) = f EV (t)·P EV,max

[0219] Among them:

[0220] PEV,demand (t) is the charging demand of the electric vehicle;

[0221] f EV (t) is the demand regulation factor based on real-time load prediction;

[0222] P EV,max is the maximum power for charging the electric vehicle.

[0223] This model is dynamically adjusted according to the actual demand to ensure the balance between the charging demand of electric vehicles and the load demand of the power station.

[0224] The present invention adopts the model predictive control algorithm mainly because MPC has powerful real-time optimization capabilities and multi-objective optimization characteristics, and has significant advantages when dealing with complex systems such as integrated photovoltaic energy storage charging and discharging power stations. By using the current system state and predicted future behavior at each control moment, MPC can dynamically schedule the operation of photovoltaic power generation, energy storage charging and discharging, and charging piles, and ensure that the power station achieves a balance among multiple objectives such as photovoltaic power consumption, grid power purchase cost, and energy storage battery life. The MPC algorithm can naturally handle multiple constraints in the operation of the power station, such as battery SOC, battery charging and discharging limits, etc., and ensure the stable and safe operation of the system. In addition, the predictability of MPC enables it to make forward-looking scheduling according to future load, weather and other external changes, thereby enhancing the system's ability to respond to environmental changes. At the same time, MPC has good flexibility and scalability, can adapt to the needs of power stations of different scales and types, and can achieve seamless connection during system upgrade and transformation. MPC can effectively handle the constraint conditions between multiple devices, such as the charge and discharge depth of the energy storage system, photovoltaic power generation fluctuations, grid power purchase price fluctuations, etc., and ensure that the system operates within the feasible region. By predicting future load, photovoltaic power generation and meteorological conditions, MPC provides real-time optimization and predictive decision-making, and can respond to system fluctuations or load changes in advance. In addition, MPC can flexibly handle multi-objective optimization problems, balance objectives such as photovoltaic power consumption rate, power purchase cost and energy storage life, and improve the economy and efficiency of the power station. Its adaptability and flexibility enable the system to adjust control parameters in real time to ensure that changes in photovoltaic fluctuations, energy storage characteristics and grid electricity price fluctuations can be effectively addressed, thereby improving the safety and reliability of the power station. The MPC algorithm can efficiently and intelligently schedule the various resources of the integrated photovoltaic energy storage charging and discharging power station through real-time optimization, handling multiple constraints, having prediction capabilities, and adapting to dynamic changes, ensuring the efficient and stable operation of the system. MPC works collaboratively at each level in the three-level control architecture of the present invention to ensure optimization and coordination at different time scales, which helps to improve the overall performance of the power station.

[0225] A state-space model is established for the energy flow characteristics of the integrated photovoltaic energy storage charging and discharging power station to describe the dynamic response of the power station system, specifically including:

[0226] x(t): State variables, including the state of charge (SOC) of the energy storage battery, photovoltaic power generation, load demand, etc.;

[0227] u(t): Control variables, including photovoltaic output regulation, energy storage charge and discharge power, charging pile scheduling plan, etc.;

[0228] y(t): Output variables, including the power purchased from the grid, current, voltage, temperature of the energy storage device, etc.;

[0229] The state - space model is in the form of:

[0230] x(t + 1)=Ax(t)+Bu(t)+Ew(t)

[0231] y(t)=Cx(t)+Du(t)

[0232] Where:

[0233] A, B, C, D are system parameter matrices;

[0234] w(t) is an external disturbance, including environmental parameters and load fluctuations.

[0235] The main purpose of presenting the state - space model is to accurately describe the dynamic behavior of the integrated photovoltaic - energy storage - charging and discharging power station and provide a mathematical basis for control algorithms (such as MPC). The state - space model can clearly express the dynamic relationships and evolution laws among various state variables of the system (such as battery SOC, photovoltaic power generation, load demand, etc.), which is crucial for implementing effective control strategies. By establishing the state - space model, it helps to analyze and understand the behavior of the system under different conditions, supports the MPC algorithm to make optimization decisions based on the dynamic model of the system. It can also effectively handle the constraint conditions in the system, such as battery SOC limits, photovoltaic power generation fluctuations, etc., providing a necessary framework for the MPC to consider these constraints when solving the optimal control strategy. In addition, the state - space model can analyze performance indicators such as the stability and response speed of the system, ensure the stable operation of the power station, and provide a basis for performance optimization. By predicting the existing state and control input, the state - space model provides accurate predictions for load demand and photovoltaic power generation fluctuations in the integrated photovoltaic - energy storage - charging and discharging power station, provides predictions of future system behavior for the MPC, and then makes reasonable scheduling decisions. Generally speaking, the state - space model is a fundamental tool for implementing the MPC algorithm, optimizing control strategies, handling constraints, analyzing system performance, and predicting scheduling.

[0236] For uncertain factors, a reinforcement learning algorithm is used to online adjust the control parameters of the model predictive control algorithm. The reinforcement learning algorithm autonomously adjusts the weight coefficients of the model predictive control algorithm through trial - and - error and feedback, dynamically optimizing the control strategy under different working conditions, as follows:

[0237] S51: Define the state space, and use the operation state information of the power station as the input state of the reinforcement learning algorithm. The state variables include:

[0238] s t =[P PV (t), SOC(t), P EV (t), T out (t), C(t)]

[0239] where

[0240] T out (t): the environmental temperature at time t;

[0241] S52: Define the action space, and use the MPC control parameters as the output actions of the reinforcement learning algorithm. The defined control parameters include:

[0242] λ t =[λ1(t), λ2(t), λ3(t), λ4(t), N(t)]

[0243] where

[0244] N(t): the rolling optimization time domain of the MPC algorithm, the control prediction time range.

[0245] S53: Define the reward function. According to the optimization goal of the integrated photovoltaic energy storage charging and discharging power station, define the reward function r t to guide the reinforcement learning algorithm to optimize the control parameters.

[0246] r t =ω1·η PV (t)-ω2·P grid (t)-ω3·DOD(t)

[0247] where:

[0248] ω1, ω2, ω3 are the weight coefficients of the reward function;

[0249] η PV (t) represents the photovoltaic accommodation rate

[0250] D battery (t) is the charge and discharge depth of the energy storage battery, indicating the discharge degree of the battery;

[0251] This reward function realizes the multi-objective trade-off of photovoltaic utilization rate, grid power purchase cost and energy storage battery health.

[0252] S54: Determine the goal of reinforcement learning, that is, find the optimal policy π * , select the optimal action at each time t to maximize the cumulative reward:

[0253]

[0254] Wherein:

[0255] π represents the policy function, defining the mapping between the state space and the action space;

[0256] γ is the discount factor, used to balance short-term rewards and long-term benefits;

[0257] r t is the immediate reward at each moment.

[0258] S55: To improve the training stability, an Experience Replay mechanism is adopted to improve the generalization ability of the model by caching historical data.

[0259] S56: Update the control parameters. In each control cycle, the reinforcement learning algorithm updates the MPC parameters according to the operating state of the power station and the prediction data. If the reward value increases, the parameters remain unchanged; if the reward value decreases, the parameter combination is adjusted to find a better solution.

[0260] Due to the volatility of photovoltaic power generation, electric vehicle charging demand, and grid electricity price, as well as the change of power station equipment performance over time, fixed control parameters are difficult to ensure the efficient and stable operation of the power station in various situations. Therefore, by real-time monitoring of environmental parameters and equipment status, and dynamically adjusting control parameters, the adaptive parameter adjustment mechanism can keep the power station in the optimal operating state all the time, improve the photovoltaic power consumption rate, reduce the electricity purchase cost, extend the energy storage life, and enhance the safety and reliability of the system.

[0261] The adaptive parameter adjustment mechanism can enhance the flexibility and operating efficiency of the integrated photovoltaic energy storage charging and discharging power station, ensuring that the system maintains an optimal state under different operating conditions. Due to the high uncertainty of load demand, photovoltaic power generation, and grid electricity prices, traditional fixed-parameter control strategies are difficult to cope with. The adaptive adjustment mechanism can dynamically optimize key parameters such as the MPC control weight and prediction window based on real-time monitoring data, improving the foresight and accuracy of optimal scheduling. At the same time, this mechanism can optimize the energy storage charging and discharging strategy, dynamically adjusting the charging and discharging behavior based on the state of health (SOH) of the battery, the number of charge-discharge cycles, and environmental factors, avoiding deep discharge and high-rate charging and discharging, thereby extending the battery life. In addition, adaptive parameter adjustment can also enhance the electricity price response ability, adjusting the charging and discharging plan in combination with electricity price prediction, charging reasonably at low electricity prices and preferentially discharging at high electricity prices to reduce operating costs. For the charging pile scheduling, this mechanism can optimize the scheduling strategy according to real-time load, electric vehicle charging demand, and power station capacity constraints, balancing the load and preventing local overload. Generally speaking, the adaptive parameter adjustment mechanism improves the photovoltaic power consumption rate, reduces the electricity purchase cost, extends the energy storage life, and enhances the charging pile scheduling efficiency by perceiving the system state in real time, flexibly adjusting control parameters, and optimizing the scheduling strategy, significantly enhancing the safety, economy, and reliability of the integrated photovoltaic energy storage charging and discharging power station.

[0262] By adopting the adaptive parameter adjustment mechanism to cope with the dynamic changes of the environment and load, improve the robustness and stability of the system, optimize the system performance, and enhance the adaptability of the system. In the present invention, the adaptive parameter adjustment is dynamically adjusted according to the power station load, environmental parameters, and equipment status. Dynamically optimize the key parameters (such as the target weight and control time domain) of the MPC algorithm to enhance the self-adaptability and robustness of the control strategy. The present invention uses a reinforcement learning algorithm to online adjust the parameters of the MPC algorithm. This method aims to improve the self-adaptability of the MPC algorithm to the power station operating environment and load fluctuations and optimize the scheduling effect.

[0263] By defining a reward function (RewardFunction): using performance indicators such as the photovoltaic power consumption rate, grid electricity purchase cost, and energy storage battery health as feedback signals for reinforcement learning; through continuous exploration and trial, the reinforcement learning algorithm can gradually optimize the MPC control parameters to achieve dynamic self-adaptation of the control strategy; when the environment changes (such as load fluctuations, weather mutations), the reinforcement learning algorithm can quickly adjust the MPC control parameters to ensure the stability and efficiency of the system response;

[0264] In the reinforcement learning algorithm, directly using the samples generated by each interaction to update the model may lead to too high data correlation, affecting the stability and convergence speed of the model. Therefore, the present invention introduces an experience replay mechanism in step S45 to improve the training efficiency, stability, and generalization ability of the algorithm.

[0265] Store the sample data (i.e., state, action, reward, and next state) obtained by the agent through interaction in the environment in an experience pool;

[0266] Each time the model is updated, randomly sample a batch of samples from the experience pool instead of using the latest data points, thereby breaking the temporal correlation between samples and enhancing the stability of training.

[0267] The steps are as follows:

[0268] S551: Initialize the experience pool and create an experience pool D with a capacity of D max ;

[0269] S552: In each control cycle, record the state (State), action (Action), reward (Reward), and next state (Next State) at the current moment to form a quadruple sample (s t , a t , r t , s t+1 ). Store this sample in the experience pool D; if the experience pool is full, dequeue the earliest sample.

[0270] Random sampling can break the temporal correlation between samples and prevent the model from falling into local optima. During the model training process, randomly draw a mini-batch of samples from the experience pool, and set the number of samples to N; update the value network according to the following objective function

[0271]

[0272] where L(θ) is the loss function;

[0273] Q′ and π′ are the Q-value function and policy function of the target network respectively.

[0274] S554: Update the Actor network parameters using policy gradient based on the sampled samples:

[0275]

[0276] Preferably, the present invention further introduces the PER mechanism to assign priorities to the samples in the experience pool. Prioritize the samples that contribute more to policy improvement, thereby accelerating the convergence speed of the model. The priority calculation formula of PER is as follows

[0277]

[0278] where,

[0279] δ i is the error of sample i

[0280] α is a hyperparameter that controls the weight of sample priority allocation.

[0281] The experience replay mechanism adopted in this invention has significant advantages. By randomly sampling to break temporal correlation, it effectively eliminates the correlation between data, significantly improving the stability of the training process. At the same time, the experience pool mechanism can make full use of historical samples, reduce the dependence on real-time data, and improve the utilization rate of samples. In the process of continuously accumulating diverse samples, the reinforcement learning model can better adapt to the operating environment of complex power plants, enhancing the generalization ability of the model. In addition, after introducing the PER mechanism, the model can focus on key samples faster, thus accelerating the optimization process of the policy and improving the overall performance.

[0282] To ensure the stability and flexibility of the system under different operating states, the control parameters are updated in real time through the following dynamic adjustment formula:

[0283]

[0284] Where:

[0285] θ opt (t) is the optimal control parameter at the current moment;

[0286] θ prev (t) is the control parameter at the previous moment;

[0287] η is the learning rate, which is used to control the update step size;

[0288] is the gradient of the control objective function with respect to the control parameter, representing the sensitivity of the control parameter to the optimization objective.

[0289] Through this adaptive adjustment mechanism, the system can optimize the control parameters in real time according to the changes in the external environment and internal state, and maintain the best operating state of the system.

[0290] In engineering applications, the intelligent control strategy of the present invention adopts a hierarchical control architecture, including the following two parts: Edge control layer: Field devices such as PLCs and embedded controllers are used to achieve direct control of photovoltaics, energy storage, and charging piles; Centralized scheduling layer: An MPC controller is deployed in the centralized monitoring platform. Combining the real-time data and prediction results of the power station, it calculates and issues optimal scheduling instructions; Through the Industrial Internet of Things (IIoT) platform, data interaction between the centralized monitoring platform and edge control devices is realized. Specifically, the control strategy of the present invention is achieved through the following means: Embed the MPC algorithm in the field controllers (such as PLCs and embedded controllers) of the integrated photovoltaic energy storage and charging and discharging power station; Transmit information such as control parameters, prediction data, and equipment status to the centralized monitoring platform in real time through the Industrial Internet of Things (IIoT) platform; In the centralized monitoring platform, online adjustment of control parameters and remote operation and maintenance are realized to ensure the long-term stable operation of the power station.

[0291] The intelligent control strategy of the present invention can be widely applied to distributed integrated photovoltaic energy storage and charging and discharging power stations in residential, commercial, and industrial parks; large charging stations and battery swapping stations to achieve collaborative optimization of photovoltaic power generation, energy storage systems, and charging piles; and microgrids and regional energy management systems to improve the intelligent level of energy scheduling.

[0292] The above are only the preferred embodiments of the present invention, and do not limit the present invention in any other form. Any modification or equivalent change made according to the technical essence of the present invention still belongs to the scope protected by the present invention.

Claims

1. An intelligent scheduling and control method for an integrated photovoltaic energy storage charging and discharging power station, characterized in that It includes the following steps: Step S1, construct a three-level control framework: Set three levels: upper, middle and lower. They are the rolling optimization layer at the upper level, the real-time scheduling layer at the middle level, and the device control layer at the lower level, Upper layer: rolling optimization layer The rolling optimization layer is responsible for performing rolling optimization within a time range. With a time resolution of 15 minutes, according to the predicted photovoltaic power generation output, load demand, energy storage charge and discharge requirements, and electricity price fluctuations, calculate the optimal control strategy within the next 24 hours. Considering multiple optimization objectives, design a multi-objective optimization function. The formula is as follows: Where: P PV (t) is the output of photovoltaic power generation; P load (t) is the power station load demand; C(t) is the electricity price at time t; SOC deviaton (t) is the deviation between the energy storage SOC and the target SOC; P storage is the real-time charge and discharge power of the energy storage device; λ1, λ2, λ3, λ4 are the weight coefficients in the objective function, which are adjusted according to the actual application situation, By optimizing this objective function, the energy storage charge and discharge strategy and photovoltaic power generation scheduling strategy within the next 24 hours can be obtained; Middle layer: real-time scheduling layer Based on the optimization results of the upper layer, the real-time scheduling layer adjusts the energy storage and photovoltaic power generation output under real-time dynamic conditions through a mathematical model. The optimization objective is to dynamically adjust the charge and discharge plan of the energy storage device, Mathematical model: Adopt the following real-time scheduling optimization model. The formula is as follows: Where: P EV (t) is the charging power of the electric vehicle at the current moment; P EV,demand (t) is the charging demand of the electric vehicle; P storage,optimal (t) is the optimal energy storage charge and discharge power obtained from the upper-layer optimization calculation; Lower layer: device control layer The device control layer realizes specific device execution through a control model, and precisely controls the energy storage device, photovoltaic power generation system and grid interface in real time, The control model is as follows: P storage,command P(t) = storage P(t)+ΔP storage (t) Where: P storage,command (t) is the control command of the energy storage device; ΔP storage (t) is the fine-tuning command provided by the middle layer control, The constraint conditions of the system are as follows Grid power purchase power constraint: P grid,min ≤P grid (t)≤P grid,max Photovoltaic power generation power constraint: 0 ≤ P PV (t) ≤ P PV,max (t) Energy storage charge and discharge power constraint: 0 ≤ P storage (t) ≤ P storage,max Energy storage SOC constraint: SOC min ≤SOC(t)≤SOC max ; Step S2, establish a prediction model Adopt a long short-term memory network model to predict the photovoltaic power generation output, load demand and electricity price at future moments; Step S3, through the prediction model control algorithm for rolling optimization, dynamically adjust the control strategy to minimize the multi-objective optimization function, specifically as follows: In the upper layer optimization, comprehensively consider the photovoltaic accommodation rate, grid power purchase cost and energy storage life, and calculate the optimal control plan within the next 24 hours; The middle layer control makes fine adjustments at the minute level according to the real-time scheduling requirements to ensure that the system operation is consistent with the real-time optimization objectives; The lower layer control is responsible for the specific execution of the device, and adjusts the output of the energy storage device and photovoltaic system through a 1-second response to meet the optimization strategy in real time; Step S4, establish a state space model to describe the dynamic response of the power station system, Specifically include: x(t): state variable, u(t): control variable, y(t): output variable, The form of the state space model is: x(t + 1) = Ax(t) + Bu(t) + Ew(t) y(t) = Cx(t) + Du(t) Where: A, B, C, D are system parameter matrices; w(t) is an external disturbance, including environmental parameters and load fluctuations; Step S5, adaptive parameter adjustment mechanism Adopt a reinforcement learning algorithm to online adjust the control parameters. The reinforcement learning algorithm autonomously adjusts the weight coefficients of the prediction model through a trial-and-error and feedback mechanism, and dynamically optimizes the control strategy under different working conditions, specifically as follows: S51: Define the state space, and use the operation state information of the power station as the input state of the reinforcement learning algorithm. The state variables include: s t = [P PV (t), SOC(t), P EV (t), T out (t), C(t)] Where T out (t): the ambient temperature at time t; S52: Define the action space, and use the control parameters of the model predictive control algorithm as the output actions of the reinforcement learning algorithm. The defined control parameters include: λ t = [λ1(t), λ2(t), λ3(t), λ4(t), N(t)] Where N(t): The rolling optimization time domain of the model predictive control algorithm, the control prediction time range; S53: Define the reward function. According to the optimization objectives of the integrated photovoltaic energy storage charging and discharging power station, define the reward function r t to guide the optimization of control parameters by the reinforcement learning algorithm r t = ω1·η PV (t) - ω2·P grid (t) - ω3·DOD(t) Where: ω1, ω2, ω3 are the weight coefficients of the reward function η PV (t) represents the PV accommodation rate D battery (t) represents the charge and discharge depth of the energy storage battery, indicating the degree of battery discharge; S54: Determine the goal of reinforcement learning, i.e., find the optimal policy π * , select the optimal action at each time step t to maximize the cumulative reward: Where: π represents the policy function, defining the mapping between the state space and the action space; γ is the discount factor, used to balance short-term rewards and long-term benefits; r t is the immediate reward for each moment; S55: Adopt an experience replay mechanism to break the temporal correlation through random sampling and eliminate the correlation between data; S56: Update the control parameters. In each control cycle, the reinforcement learning algorithm updates the control parameters of the model predictive control algorithm according to the operating state and prediction data of the power station. If the reward value increases, the control parameters remain unchanged; if the reward value decreases, the control parameter combination is adjusted to find a better solution; Step S6, Real-time feedback and adjustment Real-time monitor the operating state of power station equipment and environmental parameters. The results of real-time monitoring will be combined with the output of the model predictive control algorithm to ensure that the system can dynamically adjust the control strategy according to the changes in power station equipment parameters. When the power station equipment or environmental parameters change, the control system can quickly adjust the charge and discharge strategy of the energy storage system, the dispatching plan of photovoltaic power generation, and the decision-making of power grid power purchase according to the real-time feedback information, so as to optimize the overall operating state of the system in the shortest time.

2. The intelligent scheduling control method for an integrated photovoltaic energy storage charging and discharging power station according to claim 1, characterized in that, In step S1, the time resolution of the rolling optimization layer is 15 minutes, the time resolution of the real-time scheduling layer is 1 minute, and the time resolution of the equipment control layer is 1 second.

3. The intelligent scheduling control method for an integrated photovoltaic energy storage charging and discharging power station according to claim 2, wherein, The rolling optimization layer is a 24-hour rolling optimization.

4. The intelligent dispatching and control method for a photovoltaic energy storage charging and discharging integrated power station according to claim 1, characterized in that In step S2, the prediction models include a photovoltaic power output prediction model, an energy storage dynamic model, and a charging demand response model to achieve photovoltaic power output prediction, energy storage charge and discharge demand response, and electric vehicle charging demand response. Photovoltaic power output prediction model: Use a machine learning model based on historical data to predict the photovoltaic power generation. By training and optimizing the input of the photovoltaic power output prediction model, an accurate future photovoltaic power output prediction is obtained. The formula is as follows: P PV,pred (t) = f weather (t)·P PV,max Where: P PV,pred (t) is the predicted photovoltaic power generation f weather (t) is the light function calculated based on meteorological data P PV,max is the maximum output power of the photovoltaic system; Energy storage dynamic model: Establish a dynamic energy storage model for the energy storage system, and obtain the real-time operating state of the energy storage system through dynamic simulation. The formula is as follows: Where: E storage (t + 1) is the energy storage power at the next moment P charge (t), P discharge (t) is the charging and discharging power η charge and η discharge are the charging and discharging efficiencies, respectively; Charging demand response model: Based on the scheduling requirements of electric vehicle charging piles, design a charging demand response algorithm, and reasonably schedule the use of charging piles according to different load demands and electricity price fluctuations. The formula is as follows: P EV,demand (t) = f EV (t) · P EV,max Where: P EV,demand (t) is the charging demand of the electric vehicle f EV (t) is a demand regulation factor based on real-time load prediction P EV,max The maximum power for charging an electric vehicle.

5. The intelligent scheduling and control method for an integrated photovoltaic energy storage charging and discharging power station according to claim 1, characterized in that, In step S55, the experience replay mechanism is as follows: S551: Initialize the experience pool, create an experience pool D with a capacity of D max ; S552: In each control cycle, record the state, action, reward, and next state at the current moment to form a quadruple sample (s t , a t , r t , s t+1 ), store this sample in the experience pool D; if the experience pool is full, dequeue the earliest sample; S553: During the model training process, randomly extract a sample from the experience pool, and set the sample quantity to N; update the value network according to the following objective function Where L(θ) is the loss function; Q′ and π′ are the Q-value function and policy function of the target network respectively. S454: According to the sampled samples, use the policy gradient to update the Actor network parameters:

6. The intelligent scheduling and control method for an integrated photovoltaic energy storage charging and discharging power station according to claim 5, wherein In step S552, introduce the PER mechanism to assign priorities to the samples in the experience pool, and preferentially select the samples that contribute more to the policy improvement, so as to accelerate the convergence speed of the model. The priority calculation formula of PER is as follows Where, δ i is the error of sample i α is a hyperparameter, controlling the weight of sample priority assignment.

7. An intelligent scheduling and control method for a photovoltaic energy storage charging and discharging integrated power station according to claim 1, characterized in that, In step S56, the control parameters are updated in real time through the following dynamic adjustment formula: Where: θ opt (t) Optimal control parameter at the current moment θ prev (t) is the control parameter at the previous moment η is the learning rate, which is used to control the update step size To represent the gradient of the objective function with respect to the control parameter, which indicates the sensitivity of the control parameter to the optimization objective.

Images