Virtual power plant optimization control method based on multi-energy comprehensive scheduling

Abstract

The invention relates to a virtual power plant optimization control method based on multi-energy comprehensive scheduling, and the method comprises the steps: 1, taking cost minimization, reliability improvement and environmental influence minimization as targets, considering the power scheduling of power generation and energy storage equipment, building a virtual power plant planning optimization model through the construction of a reliability index and a target function, and carrying out the optimization of the virtual power plant planning optimization model; carrying out life cycle cost comprehensive optimization evaluation; step 2, establishing a scheduling model based on deep reinforcement learning, monitoring and controlling resources including distributed energy output, demand response load and air-conditioning load by using a multi-energy virtual power plant agent, indirectly controlling the demand response load and the air-conditioning load by adjusting the electricity price and setting the temperature, and controlling the demand response load and the air-conditioning load by using the multi-energy virtual power plant agent; therefore, load change and energy fluctuation can be flexibly handled, and electric power balance can be maintained in real time. The optimal capacity configuration problem of new energy comprehensive coordinated scheduling can be solved from the two aspects of cost and reliability.

Description

Technical Field

[0001] The present invention belongs to the technical field of virtual power plants, and in particular relates to an optimization control method for virtual power plants based on multi-energy integrated scheduling. Background Art

[0002] At present, my country's power system, which is dominated by coal-fired power, mainly adopts a deterministic approach to power balance and power generation adequacy, relying on the "source follows load" balance mode to ensure the balance of power supply and demand, and renewable energy output only serves as a supplement to the power system. With the continuous increase in the proportion of new energy, the power structure of the new power system will shift from being dominated by controllable and continuous coal-fired power generation to being dominated by new energy power generation with strong uncertainty and weak controllability. Both the supply and demand sides of the power system face strong uncertainty, and renewable energy and flexible loads need to bear part of the responsibility for power balance. The power balance mechanism will shift to a probabilistic, multi-regional, multi-subject source-grid-load-storage coordinated balance mode. In addition, power balance will highlight different contradictions at different time scales, showing the coexistence of power abandonment and power shortage risks. Therefore, virtual power plants have gradually become an important means and key force to promote the optimization and upgrading of power systems and promote the consumption of new energy.

[0003] As one of the important measures for load management, virtual power plants break the traditional boundaries between power plants and between the power generation side and the power consumption side. By aggregating and optimizing distributed power sources, energy storage equipment, controllable loads and other resources, they are formed into a whole to participate in various power market transactions. In view of the multi-level energy coordination and dispatching of the power system, virtual power plants should also build a control platform that adapts to it and design corresponding architectures and functions. At the same time, for the architecture application of this platform, it is necessary to take comprehensive energy coordination and dispatching as the goal, comprehensively utilize projects including photovoltaics, wind turbines, battery energy storage, hydrogen pools, fuel cells and other forms of energy, and solve the optimal capacity configuration problem of comprehensive coordination and dispatching of new energy from the two aspects of cost and reliability. Summary of the invention

[0004] The purpose of the present invention is to provide a virtual power plant optimization control method based on multi-energy integrated scheduling, with the goal of comprehensive energy coordinated scheduling, to comprehensively utilize projects including photovoltaics, wind turbines, battery energy storage, hydrogen tanks, fuel cells and other forms of energy, and to solve the optimal capacity configuration problem of comprehensive coordinated scheduling of new energy from the two aspects of cost and reliability.

[0005] The present invention provides a virtual power plant optimization control method based on multi-energy integrated scheduling, comprising the following steps:

[0006] Step 1: With the goal of minimizing costs, improving reliability, and minimizing environmental impact, the power dispatch of power generation and energy storage equipment is considered, and a virtual power plant planning optimization model is established by constructing reliability indicators and objective functions to conduct a comprehensive optimization evaluation of life cycle costs;

[0007] Step 2: Establish a scheduling model based on deep reinforcement learning, and use a multi-energy virtual power plant intelligent agent to monitor and control resources including distributed energy output, demand response load, and air-conditioning load. By adjusting electricity prices and set temperatures, the demand response load and air-conditioning load are indirectly controlled to flexibly respond to load changes and energy fluctuations, and maintain power balance in real time.

[0008] Furthermore, the method for establishing the virtual power plant planning optimization model in step 1 includes:

[0009] (1) Establishing a virtual power plant model:

[0010] The light radiation intensity is converted into photovoltaic output power, and the calculation formula is as follows:

[0011] P PV =G·P PVr ·η PV

[0012] Where: G is the light radiation intensity perpendicular to the photovoltaic array surface; P PVr is the rated power of the photovoltaic array; η PV is the power conversion coefficient between the DC array and the corresponding node;

[0013] The output power of a wind turbine is related to the wind speed, and the specific expression adopts the output model of a doubly-fed induction wind turbine;

[0014]

[0015] Where: Pwind, Pr are the active output power and rated active output power of the wind turbine respectively; v, v i , v r , v o , v oc , v h They are actual wind speed, cut-in wind speed, rated wind speed, cut-out wind speed, cut-out wind speed upper limit, and actual output wind speed;

[0016] The power output of a fuel cell is calculated as:

[0017] P FC =P FCt ·η FC

[0018] Where P FC is the total power of the fuel cell, ηFC for fuel cell efficiency;

[0019] The calculation formula for the hydrogen energy stored in the container is:

[0020]

[0021] Where: Δt is the reaction time; P el is the transmission power between electrolysis and hydrogen storage; η st is the efficiency of the energy storage system, including leakage and compression energy losses; P taFC Output power for hydrogen fuel cells;

[0022] Battery energy storage is used to make up for the lack of volatility in renewable energy output and to achieve a smooth power output curve through energy storage charging and discharging; the expression of its charging and discharging energy is:

[0023]

[0024] Where: E b (t) is the battery energy storage capacity; P b is the battery storage power; η bc and η bdc are the charge and discharge efficiency respectively;

[0025] (2) Select reliability indicators:

[0026] Select the expected load loss LOLE, the expected energy loss LOEE or the expected power shortage EENS, the function loss probability LPSP, and the loss factor ELF to make reliability judgment, and use the Markov chain to calculate the reliability parameters;

[0027] (3) Establish the objective function:

[0028] Taking into account the impact of cost and benefit, the objective function is defined as follows:

[0029]

[0030] Where: C NPi is the i-th device in the power grid; C NPm , C NPtd They are network maintenance cost and transaction cost respectively;

[0031] (4) Establish constraints:

[0032] At any time, the power should be balanced in real time as follows:

[0033]

[0034] Where: P L (t), P s (t), PG (t), P FC (t) are the total load demand, the upper network power, the power transmitted from the distributed generation to the battery, and the transient power stored by the fuel cell to the DC / AC converter;

[0035] Since 10% of the load in each period is the terminal load, the load interruption is not considered in the reliability calculation. When ELF is considered in the reliability calculation, the constraints are as follows:

[0036] E(I EIF )≤I EIFmax

[0037] The energy stored in the hydrogen and the energy in the battery satisfy the following constraints:

[0038] E bmin (t)≤E b (t)≤E bmax (t)

[0039] 0(t)≤E t (t)≤E tmax (t)

[0040] Unit power generation constraints:

[0041] P Gmi ≤P Gi,t ≤P Gmzx

[0042] Unit climbing constraints:

[0043] P Gi,t -P Gi,t-1,s ≤R Umaxi

[0044] P Gi,t-1 -P Gi,t ≤R Dmax

[0045] Distributed power generation output constraints:

[0046]

[0047] Furthermore, the specific expression in step (2) is as follows:

[0048]

[0049] Where: p s is the probability of state s; T s is the load loss time; N is the total number of time periods; Q s is the loss load power; D(t) is the load demand power; Q(t) is the total load.

[0050] Furthermore, in step (3):

[0051] C NPi =N i (C Ii +C Ri K i +C OMi R)

[0052]

[0053]

[0054] Where: N i is the equipment capacity; C Ii is the initial investment cost; C Ri is the replacement and maintenance cost during the project life cycle; C NPi is the initial purchase cost of hydrogen, as the electrolytic cell investment and the actual interest cost, calculated using the nominal interest rate and annual inflation rate; Y and L are equipment replacement and service life, respectively; K i , C OMi , R, r i , n are the total interest rate, operation and maintenance cost coefficient, annual interest rate, interest rate, and calculation time in the life cycle;

[0055] Load interruption cost is the cost of losses caused by power outages to users, based on the user's willingness to expand the network or calculated by the loss value of industrial products; the calculation method of annual loss expectation is as follows:

[0056]

[0057] The net present value of the loss load is calculated as:

[0058] C NPloss =I IOEE C loss R

[0059] Where: C loss is the loss load cost coefficient, which represents the average unit power loss caused by the power failure;

[0060] With the goal of minimizing operating costs, the income from selling electricity to the upper-level power grid is introduced; the net present value cost calculation formula is:

[0061]

[0062] Where: C s The revenue generated by selling electricity to the upper grid depends on the time of power exchange and the electricity price at the time of exchange.

[0063] Furthermore, the virtual power plant planning optimization model is solved using a particle swarm algorithm.

[0064] Furthermore, the method for establishing the scheduling model based on deep reinforcement learning in step 2 includes:

[0065] 1) State space definition:

[0066] Due to the existence of demand-side flexibility resources, the information observed by the multi-energy virtual power plant agent from the environment includes distributed energy output, planned output value, heat load value, SOC, demand response load size, and average room temperature of air-conditioning load users; the state space S of the multi-energy virtual power plant agent model is defined as:

[0067]

[0068] Where: is the output power of the ith distributed energy in period t; ψ(t) is the SOC of the energy storage device; E out (t) is the planned output of MEVPP in period t; H load (t) is the heat load of MEVPP; L j,DR (t) is the power of the jth demand response load in period t; L k,AC (t) is the power of the kth air conditioning load in period t; T avg (t) is the average indoor temperature of air-conditioning users; t is the current time period;

[0069] 2) Action space design:

[0070] The multi-energy virtual power plant agent selects action a in the action space A according to the observed state s; for the power generation unit or heating unit directly controlled by the multi-energy virtual power plant, its action value is the output power of the equipment For energy storage devices, the action is its charging or discharging power P ESS (t); For demand response load and air conditioning load, the multi-energy virtual power plant adjusts the real-time electricity price and sets the air conditioning temperature T AC (t) is adjusted to indirectly control the power consumed by the demand response load and the air conditioning load cluster; the corresponding actions are the electricity price adjustment range K(t), yuan / (kW·h) and the air conditioning set temperature T AC (t); the action space A is defined as:

[0071]

[0072] 3) Define constraints and reward functions:

[0073] The power balance constraints of the multi-energy virtual power plant are:

[0074]

[0075] The constraints of the multi-energy agent to meet the heat load requirements are:

[0076] Q cus (t)≤Q out (t)

[0077] Where: Q cus (t) is the heat load demand during period t; Q out (t) is the thermal power reaching the demand side during period t;

[0078] The operating cost of the multi-energy virtual power plant is defined as the objective function of the optimization scheduling problem. The reward function obtained by the agent from the environment is defined as the negative value of the objective function. According to the value of the reward function corresponding to different actions, the agent takes a series of optimal actions in each round to obtain the maximum cumulative reward, forming an optimal action sequence, that is, a scheduling scheme with the minimum operating cost. Ignoring the equipment maintenance cost and communication cost, the reward function is defined as the sum of the operating cost and penalty term of the MEVPP. The reward function R is expressed as:

[0079] R=-C ng -C ub

[0080] Where: C ng , C ub They are respectively the gas purchase cost per unit time period and the penalty item for the virtual power plant failing to meet the constraints.

[0081] Further, in step 3):

[0082] C ng (t) = c cng [M GT (t)+M GB (t)]

[0083] Where: c cng is the unit quality price of natural gas; M GT (t), M GB (t) The quality of natural gas consumed by gas turbines and gas boilers;

[0084] C ub =C ubl +C ub +C ub3 +C ub4

[0085] Where: C ubl ~C ub4 They are the penalty items for MEVPP not meeting the planned output constraint conditions, not meeting the heat load demand, not meeting the indoor temperature conditions of air-conditioning users, and the penalty items for overcharging or over-discharging of the energy storage device, among which:

[0086]

[0087] Where: C ub Penalty factor for multi-energy virtual power plants that fail to track planned output;

[0088] C ub (t) = c ub2 max[Q cus (t)-Q out (t),0]

[0089] Where: C ub2 is the penalty coefficient for the multi-energy virtual power plant failing to meet the heat load demand;

[0090]

[0091] Where: C ub T is the penalty coefficient for air-conditioning users’ indoor temperature exceeding the limit; k,AC (t) is the indoor temperature of the kth air-conditioning user in period t;

[0092] C ub (t) = c ub3 [max(0,ψ min -ψ(t))+max(0,ψ(t)-ψ max )]E max

[0093] Where: C ub4 It is the penalty coefficient for overcharging or overdischarging the energy storage device.

[0094] The present invention also provides a virtual power plant optimization control system based on multi-energy integrated scheduling, including an optimization control module, wherein the optimization control module executes the virtual power plant optimization control based on multi-energy integrated scheduling.

[0095] The present invention also provides a non-transitory computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, the virtual power plant optimization control method based on multi-energy integrated scheduling is implemented.

[0096] The present invention also provides an electronic device, comprising:

[0097] A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the virtual power plant optimization control method based on multi-energy integrated scheduling by executing the computer instructions.

[0098] Through the above scheme, the virtual power plant optimization control method based on multi-energy comprehensive scheduling has the following technical effects:

[0099] (1) The substation-level virtual power plant platform in the present invention achieves various forms of energy complementarity by aggregating various distributed energy sources (such as photovoltaics, wind turbines, battery energy storage, hydrogen energy, etc.). The platform can flexibly manage various energy forms such as electrical energy and thermal energy, thereby improving the overall efficiency of the power system and ensuring stable energy supply. Through the optimized scheduling of deep reinforcement learning, the virtual power plant intelligent body can effectively integrate and utilize various energy resources to maximize the efficiency and reliability of energy utilization.

[0100] (2) The virtual power plant platform achieves multi-objective collaborative optimization through a virtual power plant planning optimization model and a scheduling model based on deep reinforcement learning. Specifically, the system not only takes into account the power scheduling of power generation and energy storage equipment, but also takes into account the minimization of system costs, reliability improvement, and minimization of environmental impact. By constructing reliability indicators and objective functions, the platform can achieve a comprehensive optimization evaluation of life cycle costs based on initial installation costs, replacement costs, operation and maintenance costs, and environmental factors, further improving the overall performance of the energy system.

[0101] (3) Virtual power plants use precise aggregation and flexible control technologies to organically integrate distributed power sources, energy storage equipment, and controllable loads. Through the control platform, the platform can maintain power balance in real time, and use multi-energy virtual power plant intelligent entities to monitor and control resources including distributed energy output, demand response loads, and air conditioning loads. By adjusting electricity prices and set temperatures, indirect control of demand response loads and air conditioning loads is achieved, thereby flexibly responding to load changes and energy fluctuations, improving the stability and economy of the system.

[0102] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 It is a flow chart of the virtual power plant optimization control method based on multi-energy integrated scheduling of the present invention;

[0104] Figure 2 It is the main framework for multi-level energy interaction;

[0105] Figure 3 It is a schematic diagram of the structure of the electronic device of the present invention. DETAILED DESCRIPTION

[0106] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0107] Ginseng Figure 1 As shown, this embodiment provides a virtual power plant optimization control method based on multi-energy integrated scheduling, including the following steps:

[0108] Step S1, with the goal of minimizing costs, improving reliability and minimizing environmental impact, considering the power dispatch of power generation and energy storage equipment, establishing a virtual power plant planning optimization model by constructing reliability indicators and objective functions, and conducting a comprehensive optimization evaluation of life cycle costs;

[0109] Step S2, establish a scheduling model based on deep reinforcement learning, use a multi-energy virtual power plant intelligent agent to monitor and control resources including distributed energy output, demand response load, and air-conditioning load, and indirectly control the demand response load and air-conditioning load by adjusting the electricity price and setting the temperature, so as to flexibly respond to load changes and energy fluctuations and maintain power balance in real time.

[0110] The present invention is described in further detail below.

[0111] The optimization control method of virtual power plants based on multi-energy integrated scheduling mainly includes two aspects: the construction of virtual power plant planning optimization model based on the control platform and the optimization scheduling of multi-energy virtual power plants based on deep reinforcement learning.

[0112] a) Virtual power plant planning optimization model based on control platform

[0113] Strp1. Establish a virtual power plant model

[0114] Figure 2 The main architecture of multi-level energy interaction in the market framework is given. In the autonomous microgrid, distributed generation can provide power with sufficient capacity and can also realize island operation with the upper network.

[0115] The light radiation intensity is converted into photovoltaic output power, and the calculation formula is as follows:

[0116] P PV =G·P PVr ·η PV

[0117] Where: G is the light radiation intensity perpendicular to the photovoltaic array surface; P PVr is the rated power of the photovoltaic array; η PV is the power conversion coefficient between the DC array and the corresponding node.

[0118] The output power of a wind turbine is related to the wind speed, and the specific expression adopts the output model of a doubly-fed induction wind turbine.

[0119]

[0120] Where: Pwind and Pr are the active output power and rated active output power of the wind turbine respectively; v, vi, vr, vo, voc and vh are the actual wind speed, cut-in wind speed, rated wind speed, cut-out wind speed, cut-out wind speed upper limit and actual output wind speed respectively.

[0121] The power output of a fuel cell is calculated as:

[0122] P FC =P FCt ·η FC

[0123] Where P FCt is the total power of the fuel cell, η FC for fuel cell efficiency.

[0124] The calculation formula for the hydrogen energy stored in the container is:

[0125]

[0126] Where: Δt is the reaction time; P el is the transmission power between electrolysis and hydrogen storage; η st is the efficiency of the energy storage system, including leakage and compression energy losses; P taFC Output power for hydrogen fuel cells.

[0127] Battery energy storage is used to make up for the volatility of renewable energy output and other deficiencies, and to achieve the purpose of smoothing the power output curve through energy storage charging and discharging. The expression of its charging and discharging energy is:

[0128]

[0129] Where: E b (t) is the battery energy storage capacity; P b is the battery storage power; η bc and η bdc are the charge and discharge efficiency respectively.

[0130] Step 1.2 Select reliability indicators

[0131] Select the expected loss of load (LOLE), expected loss of energy (LOEE) or expected energy shortage (EENS), loss of function probability (LPSP), and loss factor (ELF) for reliability judgment, and use the Markov chain to calculate the reliability parameters. The specific expression is as follows.

[0132]

[0133] Where: p s is the probability of state s; T s is the load loss time; N is the total number of time periods; Q s is the loss load power; D(t) is the load demand power; Q(t) is the total load.

[0134] Step 1.3 Establish the objective function

[0135] Life cycle cost analysis evaluates all costs over the life of the equipment. Net present value costs include initial installation costs, replacement costs, repair and maintenance costs, failure costs, and networking costs. In the calculation process, costs are all positive and benefits are all negative. During the evaluation period, all costs and expenses are calculated using a fixed interest rate. In this evaluation method, in order to increase the calculation rate at the end of the analysis process, the net present value cost should be used and the fluctuation of the real interest rate affected by inflation should be considered. The real interest rate is the difference between the nominal interest rate and the inflation rate. The formula for calculating the net present value cost of the equipment is as follows.

[0136] C NPi =N i (C Ii +C Ri K i +C OMi R)

[0137]

[0138]

[0139] Where: N i is the equipment capacity; C Ii is the initial investment cost; C Ri is the replacement and maintenance cost during the project life cycle; C NPi is the initial purchase cost of hydrogen, as the electrolytic cell investment and the actual interest cost, calculated using the nominal interest rate and annual inflation rate; Y and L are equipment replacement and service life, respectively; K i , C OMi , R, r i , n are the total interest rate during the life cycle, operation and maintenance cost coefficient, annual interest rate, interest rate, and calculation time period.

[0140] Load interruption costs are the costs of causing power outages to customers. There are different ways to measure this cost. For example, it can be based on the customer's willingness to expand the network or calculated from the value of industrial product losses. The annual loss expectation is calculated as follows:

[0141]

[0142] The net present value of the loss load is calculated as:

[0143] C NPloss =I IOEE C loss R

[0144] Where: C loss is the loss load cost coefficient, which represents the average unit power loss caused by the power failure.

[0145] Since the goal is to minimize operating costs, the income from selling electricity to the upper grid is introduced. The net present value cost calculation formula is:

[0146]

[0147] Where: C s The revenue generated by selling electricity to the upper grid depends on the time of power exchange and the electricity price at the time of exchange.

[0148] Taking into account the impact of cost and benefit, the objective function is defined as follows:

[0149]

[0150] Where: C NPi is the i-th device in the power grid; C NPm , C NPtd They are network maintenance cost and transaction cost respectively.

[0151] Step 1.4 Establish constraints

[0152] At any time, the power should be balanced in real time as follows:

[0153]

[0154] Where: P L (t), P s (t), P G (t), P FC (t) are the total load demand (interruptible load and non-interruptible load), the upper network power, the power transmitted from the distributed generation to the battery, and the transient power stored by the fuel cell to the DC / AC converter.

[0155] Since 10% of the load in each period is the terminal load, the interruption of the load can be ignored in the reliability calculation. However, ELF needs to be considered in the reliability calculation, and its constraints are as follows:

[0156] E(I EIF )≤I EIFmax

[0157] The energy stored in the hydrogen and the energy in the battery should meet the following constraints:

[0158] E bmin (t)≤E b (t)≤E bmax (t)

[0159] 0(t)≤E t (t)≤E tmax (t)

[0160] Unit power generation constraints:

[0161] P Gmin ≤P Gi,t ≤P Gmaxi

[0162] Unit climbing constraints:

[0163] P Gi,t -P Gi,t-1,s ≤R Umaxi

[0164] P Gi,t-1 -P Gi,t ≤R Dmaxi

[0165] Distributed power generation output constraints:

[0166]

[0167] The above model is solved using particle swarm algorithm.

[0168] b) Optimal scheduling of multi-energy virtual power plants based on deep reinforcement learning

[0169] Virtual power plants can play the same role as traditional power plants in the operation of the power system and reduce the negative impact of high-proportion access to distributed energy on the power grid. Multi-energy virtual power plants further aggregate distributed energy in various forms, including electrical energy and thermal energy, to achieve multi-energy complementarity of different internal energy resources, thereby participating in the operation of the power system more efficiently.

[0170] Step 1.1 State space definition

[0171] Due to the existence of demand-side flexibility resources, the information observed by the multi-energy virtual power plant agent from the environment includes not only distributed energy output, planned output value, thermal load value, but also SOC, demand response load size, average room temperature of air-conditioning load users, etc. Therefore, the state space S of the multi-energy virtual power plant agent model is defined as:

[0172]

[0173] Where: is the output power of the ith distributed energy in period t; ψ(t) is the SOC of the energy storage device; E out (t) is the planned output of MEVPP in period t; H load (t) is the heat load of MEVPP; L j,DR (t) is the power of the jth demand response load in period t; L k,AC (t) is the power of the kth air conditioning load in period t; T avg (t) is the average indoor temperature of air-conditioning users; t is the current time period.

[0174] Step 1.2 Action Space Design

[0175] The multi-energy virtual power plant agent selects action a in the action space A according to the observed state s. For the power generation unit or heating unit that can be directly controlled by the multi-energy virtual power plant, the action value is the output power of the equipment For energy storage devices, the action is its charging or discharging power P ESS (t). For demand response load and air conditioning load, the multi-energy virtual power plant cannot directly control them, but adjusts the real-time electricity price and sets the air conditioning temperature T AC (t) is adjusted to indirectly control the power consumed by the demand response load and the air conditioning load cluster. The corresponding actions are the electricity price adjustment range K(t), yuan / (kW·h) and the air conditioning set temperature T AC (t); therefore the action space A is defined as:

[0176]

[0177] Step 1.3 Define constraints and reward function

[0178] The power balance constraints of the multi-energy virtual power plant are:

[0179]

[0180] The constraints of the multi-energy agent to meet the heat load requirements are:

[0181] Q cus (t)≤Q out (t)

[0182] Where: Q cus (t) is the heat load demand during period t; Q out (t) is the thermal power reaching the demand side during period t.

[0183] If the operating cost of the multi-energy virtual power plant is defined as the objective function of the optimization scheduling problem, the reward function obtained by the agent from the environment can be defined as the negative value of the objective function. According to the value of the reward function corresponding to different actions, the agent takes a series of optimal actions in each round to obtain the maximum cumulative reward, forming an optimal action sequence, that is, a scheduling scheme with the lowest operating cost. Ignoring the equipment maintenance cost and communication cost, the reward function is defined as the sum of the operating cost and penalty term of the MEVPP. The reward function R can be expressed as:

[0184] R=-C ng -C ub

[0185] Where: C ng , C ub They are respectively the gas purchase cost per unit time period and the penalty item for the virtual power plant failing to meet the constraints.

[0186] C ng (t) = c cng [M GT (t)+M GB (t)]

[0187] Where: c cng is the price per unit mass of natural gas; M GT (t), M GB (t) is the mass of natural gas consumed by gas turbines and gas boilers.

[0188] C ub =C ubl +C ub +C ub3 +C ub

[0189] Where: C ubl ~C ub4 They are the penalty items for MEVPP not meeting the planned output constraint conditions, not meeting the heat load demand, not meeting the indoor temperature conditions of air-conditioning users, and the penalty items for overcharging or over-discharging of the energy storage device, among which:

[0190]

[0191] Where: C ub Penalty factor for multi-energy virtual power plants that fail to track planned output.

[0192] C ub (t) = c ub max[Q cus (t)-Q out (t),0]

[0193] Where: c ub2It is the penalty coefficient for the multi-energy virtual power plant failing to meet the heat load demand.

[0194]

[0195] Where: C ub3 T is the penalty coefficient for air-conditioning users’ indoor temperature exceeding the limit; k,AC (t) is the indoor temperature of the kth air-conditioning user in period t.

[0196] C ub4 (t) = c ub3 [max(0,ψ min -ψ(t))+max(0,ψ(t)-ψ max )]E max

[0197] Where: c ub4 It is the penalty coefficient for overcharging or overdischarging the energy storage device.

[0198] This embodiment also provides a virtual power plant optimization control system based on multi-energy integrated scheduling, including an optimization control module, and the optimization control module executes the virtual power plant optimization control method based on multi-energy integrated scheduling.

[0199] This embodiment also provides a non-transitory computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, they implement the virtual power plant optimization control method based on multi-energy integrated scheduling.

[0200] Ginseng Figure 3 As shown, this embodiment also provides an electronic device, including:

[0201] A memory 201 and a processor 202, wherein the memory 201 and the processor 202 are communicatively connected to each other, the memory 201 stores computer instructions, and the processor 202 executes the computer instructions to execute the virtual power plant optimization control method based on multi-energy integrated scheduling.

[0202] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A virtual power plant optimization control method based on multi-energy integrated scheduling, characterized in that: The steps include: Step 1: With the goal of minimizing costs, improving reliability, and minimizing environmental impact, the power dispatch of power generation and energy storage equipment is considered, and a virtual power plant planning optimization model is established by constructing reliability indicators and objective functions to conduct a comprehensive optimization evaluation of life cycle costs; Step 2: Establish a scheduling model based on deep reinforcement learning, and use a multi-energy virtual power plant intelligent agent to monitor and control resources including distributed energy output, demand response load, and air-conditioning load. By adjusting electricity prices and set temperatures, the demand response load and air-conditioning load are indirectly controlled to flexibly respond to load changes and energy fluctuations, and maintain power balance in real time.

2. The virtual power plant optimization control method based on multi-energy integrated scheduling according to claim 1 is characterized in that: The method for establishing the virtual power plant planning optimization model in step 1 includes: (1) Establishing a virtual power plant model: The light radiation intensity is converted into photovoltaic output power, and the calculation formula is as follows: P.S PV G·P PVr ·η PV Where: G is the light radiation intensity perpendicular to the photovoltaic array surface; P PVr is the rated power of the photovoltaic array; η PV is the power conversion coefficient between the DC array and the corresponding node; The output power of a wind turbine is related to the wind speed, and the specific expression adopts the output model of a doubly-fed induction wind turbine; Where: Pwind, Pr are the active output power and rated active output power of the wind turbine respectively; v, v i , v r , v o , v oc , v h They are actual wind speed, cut-in wind speed, rated wind speed, cut-out wind speed, cut-out wind speed upper limit, and actual output wind speed; The power output of a fuel cell is calculated as: P.S FC JP FC ·η FC Where P FCt is the total power of the fuel cell, η FC for fuel cell efficiency; The calculation formula for the hydrogen energy stored in the container is: Where: Δt is the reaction time; P el is the transmission power between electrolysis and hydrogen storage; η st is the efficiency of the energy storage system, including leakage and compression energy losses; P taFC Output power for hydrogen fuel cells; Battery energy storage is used to make up for the lack of volatility in renewable energy output and to achieve a smooth power output curve through energy storage charging and discharging; the expression of its charging and discharging energy is: Where: E b (t) is the battery energy storage capacity; P b is the battery storage power; η bc and η bdc are the charge and discharge efficiency respectively; (2) Select reliability indicators: Select the expected load loss LOLE, the expected energy loss LOEE or the expected power shortage EENS, the function loss probability LPSP, and the loss factor ELF to make reliability judgment, and use the Markov chain to calculate the reliability parameters; (3) Establish the objective function: Taking into account the impact of cost and benefit, the objective function is defined as follows: Where: C NP is the i-th device in the power grid; C NP , C NPtd They are network maintenance cost and transaction cost respectively; (4) Establish constraints: At any time, the power should be balanced in real time as follows: Where: P L (t), P s (t), P G (t), P FC (t) are the total load demand, the upper network power, the power transmitted from the distributed generation to the battery, and the transient power stored by the fuel cell to the DC / AC converter; Since 10% of the load in each period is the terminal load, the load interruption is not considered in the reliability calculation. When ELF is considered in the reliability calculation, the constraints are as follows: E(I EIF )≤I EIFmax The energy stored in the hydrogen and the energy in the battery satisfy the following constraints: E bmin (t)≤E b (t)≤E bmax (t) 0(t)≤E t (t)≤E tmax (t) Unit power generation constraints: P Gmini ≤P Gi,t ≤P Gmzxi Unit climbing constraints: P Gi,t -P Gi,t-1,s ≤R Umaxi P Gi,t-1 -P Gi,t ≤R Dmaxi Distributed power generation output constraints:

3. The virtual power plant optimization control method based on multi-energy integrated scheduling according to claim 2 is characterized in that: The specific expression in step (2) is as follows: Where: p s is the probability of state s; T s is the load loss time; N is the total number of time periods; Q s is the loss load power; D(t) is the load demand power; Q(t) is the total load.

4. The virtual power plant optimization control method based on multi-energy integrated scheduling according to claim 3 is characterized in that: In step (3): C NPi =N i (C Ii +C Ri K i +C OMi R) Where: N i is the equipment capacity; C Ii is the initial investment cost; C Ri is the replacement and maintenance cost during the project life cycle; C NPi is the initial purchase cost of hydrogen, as the electrolytic cell investment and the actual interest cost, calculated using the nominal interest rate and annual inflation rate; Y and L are equipment replacement and service life, respectively; K i , C OMi , R, r i , n are the total interest rate, operation and maintenance cost coefficient, annual interest rate, interest rate, and calculation time in the life cycle; Load interruption cost is the cost of losses caused by power outages to users, based on the user's willingness to expand the network or calculated by the loss value of industrial products; the calculation method of annual loss expectation is as follows: The net present value of the loss load is calculated as: C NPloss =I IOEE C loss R Where: C loss is the loss load cost coefficient, which represents the average unit power loss caused by the power failure; With the goal of minimizing operating costs, the income from selling electricity to the upper-level power grid is introduced; the net present value cost calculation formula is: Where: C s The revenue generated by selling electricity to the upper grid depends on the time of power exchange and the electricity price at the time of exchange.

5. The virtual power plant optimization control method based on multi-energy integrated scheduling according to claim 4 is characterized in that: The virtual power plant planning optimization model is solved using a particle swarm algorithm.

6. The virtual power plant optimization control method based on multi-energy integrated scheduling according to claim 5 is characterized in that: The method for establishing the scheduling model based on deep reinforcement learning described in step 2 includes: 1) State space definition: Due to the existence of demand-side flexibility resources, the information observed by the multi-energy virtual power plant agent from the environment includes distributed energy output, planned output value, heat load value, SOC, demand response load size, and average room temperature of air-conditioning load users; the state space S of the multi-energy virtual power plant agent model is defined as: Where: G i,DER (t) is the output power of the ith distributed energy in period t; ψ(t) is the SOC of the energy storage device; E out (t) is the planned output of MEVPP in period t; H load (t) is the heat load of MEVPP; L j,DR (t) is the power of the jth demand response load in period t; L k,AC (t) is the power of the kth air conditioning load in period t; T avg (t) is the average indoor temperature of air-conditioning users; t is the current time period; 2) Action space design: The multi-energy virtual power plant agent selects action a in the action space A according to the observed state s; for the power generation unit or heating unit directly controlled by the multi-energy virtual power plant, its action value is the output power G of the equipment. i,DER (t); For energy storage devices, its action is its charging or discharging power P ESS (t); For demand response load and air conditioning load, the multi-energy virtual power plant adjusts the real-time electricity price and sets the air conditioning temperature T AC (t) is adjusted to indirectly control the power consumed by the demand response load and the air conditioning load cluster; the corresponding actions are the electricity price adjustment range K(t), yuan / (kW·h) and the air conditioning set temperature T AC (t); the action space A is defined as: 3) Define constraints and reward functions: The power balance constraints of the multi-energy virtual power plant are: The constraints of the multi-energy agent to meet the heat load requirements are: Q cus (t)≤Q out (t) Where: Q cus (t) is the heat load demand during period t; Q out (t) is the thermal power reaching the demand side during period t; The operating cost of the multi-energy virtual power plant is defined as the objective function of the optimization scheduling problem. The reward function obtained by the agent from the environment is defined as the negative value of the objective function. According to the value of the reward function corresponding to different actions, the agent takes a series of optimal actions in each round to obtain the maximum cumulative reward, forming an optimal action sequence, that is, a scheduling scheme with the minimum operating cost. Ignoring the equipment maintenance cost and communication cost, the reward function is defined as the sum of the operating cost and penalty term of the MEVPP. The reward function R is expressed as: R=-C ng -C ub Where: C ng , C ub They are respectively the gas purchase cost per unit time period and the penalty item for the virtual power plant failing to meet the constraints.

7. The virtual power plant optimization control method based on multi-energy integrated scheduling according to claim 6 is characterized in that: In step 3): C ng (t)=c cng [M GT (t)+M GB (t)] Where: c cng is the price per unit mass of natural gas; M GT (t), M GB (t) The quality of natural gas consumed by gas turbines and gas boilers; C ub =C ub1 +C ub +C ub3 +C ub4 Where: C ub1 ~Cu b4 They are the penalty items for MEVPP not meeting the planned output constraint conditions, not meeting the heat load demand, not meeting the indoor temperature conditions of air-conditioning users, and the penalty items for overcharging or over-discharging of the energy storage device, among which: Where: C ub1 Penalty factor for multi-energy virtual power plants that fail to track planned output; C ub2 (t)=c ub max[Q cus (t)-Q out (t),0] Where: C ub2 is the penalty coefficient for the multi-energy virtual power plant failing to meet the heat load demand; Where: C ub3 T is the penalty coefficient for air-conditioning users’ indoor temperature exceeding the limit; k,AC (t) is the indoor temperature of the kth air-conditioning user in period t; C ub4 (t)=c ub3 [max(0,ψ min -ψ(t))+max(0,ψ(t)-ψ max )]E max Where: C ub4 It is the penalty coefficient for overcharging or overdischarging the energy storage device.

8. A virtual power plant optimization control system based on multi-energy integrated scheduling, characterized in that: It includes an optimization control module, which executes a virtual power plant optimization control based on multi-energy integrated scheduling as described in any one of claims 1-7.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, which, when executed by a processor, implement a virtual power plant optimization control method based on multi-energy integrated scheduling as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the virtual power plant optimization control method based on multi-energy integrated scheduling as described in any one of claims 1 to 7 by executing the computer instructions.

Images