Multi-type energy collaborative secondary frequency modulation distributed optimization control method considering carbon emission

Abstract

The invention discloses a multi-type energy collaborative secondary frequency modulation distributed optimization control method considering carbon emission. The method comprises the following steps: establishing mathematical models of an energy storage system, pumped storage and a virtual power plant; on the basis of model prediction control, state evolution of a secondary frequency modulation system participated by multiple types of energy is predicted in a prediction domain, an energy storage charge state evolution prediction model and a temperature control load temperature evolution prediction model in a virtual power plant are established, and a rolling optimization mechanism is implemented; a multi-type energy collaborative frequency modulation centralized optimization model considering carbon emission is constructed, distributed solution is realized in combination with an alternating direction multiplier method, and unified optimization of economy, sustainability, environmental protection and reliability is realized; non-linear constraints are processed through a large M method, and solvability of a complex optimization problem is guaranteed; distributed optimization solution is realized by adopting an alternating direction multiplier method, and a large-scale centralized problem is decomposed into a plurality of node sub-problems to be solved in parallel, so that the calculation efficiency and the system expandability are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of automatic control of power systems, and in particular to a multi-type energy collaborative grid secondary frequency modulation optimization control method based on model predictive control and distributed optimization and considering carbon emissions, which is particularly suitable for large-scale grid secondary frequency modulation coordination control of a large-scale grid containing multiple nodes, each node being configured with energy storage systems, pumped storage and virtual power plants and other energy resources. BACKGROUND

[0002] With the increasing penetration of renewable energy in power systems, the installed capacity of new energy power generation such as wind power and photovoltaic power has shown explosive growth. However, the inherent intermittent, random and volatile characteristics of renewable energy have brought unprecedented challenges to the frequency stability control of power systems. Traditional power systems mainly rely on thermal power units to provide frequency modulation services, but thermal power units have slow response speed, low regulation accuracy and serious environmental pollution, which have made it difficult to meet the urgent demand for high-quality, fast-response frequency modulation services in high-proportion renewable energy power systems.

[0003] Grid frequency stability is a basic requirement for safe operation of power systems. Frequency deviation not only affects the normal operation of power equipment, but also may cause large-scale power outages. As an important part of grid frequency control, secondary frequency modulation requires the elimination of frequency deviation and the restoration of system frequency to the rated value within a minute time scale. Unlike local automatic response of primary frequency modulation, secondary frequency modulation requires centralized coordinated control through an automatic generation control system, which provides a technical basis for the collaborative optimization of multi-type energy and the application of advanced control algorithms.

[0004] In recent years, new frequency modulation resources such as energy storage systems, pumped storage and virtual power plants have gradually become an important supplement to grid frequency modulation services due to their unique technical advantages. Energy storage systems have the advantages of fast response speed, high regulation accuracy and strong bidirectional regulation capability, and can respond to frequency modulation instructions within seconds, but their state-of-charge dynamic variation characteristics require fine state management and forward-looking planning. Pumped storage has the characteristics of large capacity, mature technology and low operating cost, and is suitable for large-capacity, long-time frequency modulation tasks, but its start-stop and mode switching have time delay, which requires advance prediction and reasonable scheduling. Virtual power plants aggregate distributed energy resources such as temperature-controlled loads and electric vehicles to form a large-scale frequency modulation capability, and have the advantages of abundant resources, moderate cost and environmental friendliness, but their frequency modulation potential is affected by many factors such as environmental temperature and user behavior, showing significant time variation and uncertainty. These new frequency modulation resources have strong complementary technical characteristics, but their dynamic evolution characteristics and constraint conditions are complex, providing a good foundation for building a multi-type energy collaborative frequency modulation system while also requiring higher intelligence and refinement of control strategies.

[0005] However, existing frequency control techniques still face many technical bottlenecks and challenges when dealing with complex scenarios involving multiple types of energy collaborative frequency modulation. First, there is a lack of forward-looking optimization capability. Traditional frequency control methods are mainly based on current system state for immediate feedback control, lacking prediction and forward-looking planning of future system state evolution trajectory. This passive response mode often leads to improper allocation of frequency modulation resources and low system operation efficiency when facing complex time-varying characteristics such as energy storage state of charge changes, virtual power plant temperature control load thermal inertia evolution, and pumped storage reservoir dynamics. Model predictive control, as an advanced control method, can effectively handle system multivariable and multi-constraint characteristics by optimizing future control sequences within a limited prediction domain, achieving forward-looking optimization. However, existing frequency control techniques rarely use such advanced control methods.

[0006] Second, multi-objective optimization coordination is difficult. In actual operation, multiple mutually restrictive objectives such as frequency modulation economy, frequency modulation reliability, device sustainability, and environmental impact need to be considered. Traditional methods often use single-objective optimization or simple weighting, making it difficult to find a reasonable balance between multiple objectives. Especially under multiple time scales, the response characteristics of different frequency modulation resources differ significantly, and there is a lack of unified optimization framework to handle such spatio-temporal coupling.

[0007] Third, the ability to handle complex constraints is insufficient. Multiple types of energy participating in frequency modulation involve power balance constraints, device physical constraints, operation safety constraints, and state evolution constraints, which have strong coupling and dynamic time-varying characteristics. Traditional constraint handling methods have low solving efficiency and poor convergence when facing complex constraints such as energy storage SOC dynamic evolution constraints, virtual power plant temperature control load nonlinear switch constraints, and pumped storage multi-mode operation constraints.

[0008] Fourth, large-scale distributed system solving ability is insufficient. With the continuous expansion of the power system and the large-scale access of distributed energy, traditional centralized optimization methods face problems such as exponential growth of computational complexity, heavy communication burden, and single-point failure risk, making it difficult to meet the real-time control needs of large-scale distributed systems. Distributed optimization algorithms such as the alternating direction method can decompose large-scale problems into multiple sub-problems for parallel solving, significantly improving computational efficiency and system scalability, but their application in the field of multiple types of energy collaborative frequency modulation is relatively rare.

[0009] More importantly, as carbon emission constraints become increasingly stringent, the operation of the power system must balance the dual requirements of economy and environmental protection. Traditional frequency regulation optimization models mainly focus on cost minimization and do not adequately consider the impact of carbon emissions, making it difficult to adapt to the development needs of low-carbon power systems. At the same time, the state of charge of the energy storage system will change during the provision of frequency regulation services. If there is a lack of reasonable prediction and management of future state trajectories, the energy storage system may not be able to continuously provide frequency regulation services, and even the service life of the equipment may be affected due to overcharging and overdischarging. The temperature-controlled load in the virtual power plant has thermal inertia characteristics, and its temperature state evolution and frequency regulation capability exhibit a significant time delay effect, requiring forward-looking scheduling based on a prediction model. Existing technologies lack systematic consideration and advanced control methods in these dynamic state management aspects, making it difficult to achieve sustainable and efficient utilization of multiple types of energy resources.

[0010] In addition, when the frequency regulation demand exceeds the total capacity of available resources or equipment fails, traditional control methods lack effective risk control mechanisms and emergency response strategies, which may lead to frequency regulation failure and affect the stability of the grid frequency. How to establish an effective frequency regulation failure prevention and control mechanism based on the model predictive control framework to improve the robustness of the system through forward-looking optimization is a key technical problem to ensure the reliability of secondary frequency regulation. SUMMARY

[0011] The purpose of the present application is to overcome the shortcomings and deficiencies of the prior art, and to provide a multi-type energy collaborative secondary frequency regulation distributed optimization control method considering carbon emissions. This method establishes mathematical models of energy storage systems, pumped storage, virtual power plants and other frequency regulation resources, constructs a comprehensive objective function including generation cost, frequency regulation resource operation cost, energy storage state of charge deviation cost, carbon emission cost and frequency regulation failure penalty cost, and realizes distributed solving by combining the alternating direction multiplier method to achieve unified optimization of economy, sustainability, environmental protection and reliability. The large M method is used to handle nonlinear constraints to ensure the solvability of complex optimization problems. The alternating direction multiplier method is used to realize distributed optimization solving, which decomposes large-scale centralized problems into multiple node sub-problems for parallel solving to significantly improve computational efficiency and system scalability.

[0012] To achieve the above purpose, the technical scheme adopted by the present application is as follows: a multi-type energy collaborative secondary frequency regulation distributed optimization control method considering carbon emissions, comprising:

[0013] Step 1, set initial parameters, including technical parameters and operation constraints of energy storage systems, pumped storage and virtual power plants, power grid topology and line parameters, prediction domain length and control domain length of model predictive control; collect real-time power demand of each node of the power grid, state of charge of energy storage, indoor temperature of temperature-controlled load and weather forecast data through the secondary frequency regulation system of multiple types of energy;

[0014] Step 2, establish a mathematical model of energy storage system, pumped storage power station and virtual power plant participating in secondary frequency modulation, considering the dynamic characteristics, physical constraints and operation limits of various frequency modulation resources;

[0015] Step 3, based on model predictive control, predict the state evolution of the secondary frequency modulation system participated by multiple types of energy in the prediction domain, and establish a state of charge evolution prediction model of energy storage system and a temperature evolution prediction model of temperature-controlled load in virtual power plant;

[0016] Step 4, implement a rolling optimization mechanism for the evolution prediction model established in step 3, execute only the control action at the first time of the optimal control sequence in each control period, and re-solve the optimization problem based on the updated state information of the secondary frequency modulation system participated by multiple types of energy in the next control period to realize dynamic feedback correction and real-time adaptation;

[0017] Step 5, build a centralized optimization model of multiple types of energy collaborative frequency modulation considering carbon emissions, aiming to minimize system operation cost, carbon emission cost and frequency modulation failure penalty cost, while meeting power balance constraints, generator constraints and power flow constraints;

[0018] Step 6, convert the nonlinear working state constraints of temperature-controlled load in virtual power plant into mixed integer linear constraints by using the big M method, and establish a linear relationship between temperature indicator variables and air conditioner operating state by introducing auxiliary binary variables;

[0019] Step 7, based on the alternating direction multiplier method, decompose the centralized optimization problem into multiple local optimization problems according to nodes, solve the local optimization problems independently at each node, and realize global coordination through auxiliary variable updating and Lagrange multiplier updating until the optimal solution is converged.

[0020] Further, in step 2), the mathematical model of energy storage system participating in secondary frequency modulation includes the SOC dynamic evolution equation of energy storage system, power constraint, SOC constraint and power change rate constraint; as a frequency modulation resource with fast response speed and high regulation accuracy, the SOC dynamic evolution process of energy storage system directly affects its continuous frequency modulation capability;

[0021] The mathematical model of pumped storage power station participating in secondary frequency modulation includes the power constraint, upper and lower reservoir capacity constraint and power change rate constraint of pumped storage power station;

[0022] The mathematical model of virtual power plant participating in secondary frequency modulation includes temperature dynamic change equation and working state constraint.

[0023] Further, the model predictive control defines the prediction domain length as N p , the control domain length as N c , and N c ≤N pAt the current time t, the model predictive controller needs to predict the multi-type energy participation in the secondary frequency modulation system state evolution from time t+1 to t+N p and optimize the control sequence from time t to t+N c -1.

[0024] Further, in step 3, the energy storage state of charge evolution prediction model is represented as:

[0025]

[0026] Wherein, k=1, 2, …, N p , S n,i (t+k|t) represents the SOC value of the i th energy storage unit of node n at time t+k predicted based on the information at time t, P storage,n,i (t+k-1|t) represents the power output of the i th energy storage unit of node n at time t+k-1 predicted based on the information at time t, E cap,n,i represents the rated capacity of the i th energy storage unit of node n, and Δt represents the time interval.

[0027] Further, in step 3, the temperature evolution prediction model of the virtual power plant temperature-controlled load is represented as:

[0028]

[0029] Wherein, T in,n,k,j (t+k-1|t) represents the indoor temperature of the j th air conditioner in the k th virtual power plant of node n at time t+k-1 predicted based on the information at time t, Δt represents the time interval, T out,n,k,j (t+k-1|t) represents the outdoor temperature of the j th air conditioner in the k th virtual power plant of node n at time t+k-1 predicted based on the information at time t, C n,k,j represents the equivalent heat capacity of the j th air conditioner in the k th virtual power plant of node n, R n,k,j represents the equivalent thermal resistance of the j th air conditioner in the k th virtual power plant of node n, and Q n,k,j (t+k-1|t) represents the refrigeration / heat of the j th air conditioner in the k th virtual power plant of node n at time t+k predicted based on the information at time t.

[0030] Further, in step 4, in order to consider the prediction uncertainty, a rolling optimization mechanism is introduced in the prediction domain, and in each control period, the following finite time domain optimization problem is solved:

[0031]

[0032] Where J(t+k|t) represents the instantaneous cost at time t+k, ΔU(t+k|t) represents the control increment, ρ1 is the control increment weight, and J terminal (t+N p |t) is the terminal cost function, and ρ2 is the terminal weight; the real-time operating cost includes power generation cost, frequency regulation resource operating cost, energy storage SOC deviation cost, carbon emission cost, and frequency regulation failure penalty cost;

[0033] The control increment is defined as:

[0034] ΔU(t+k|t)=U(t+k|t)-U(t+k-1|t)

[0035] Where, U(t+k|t)=[P st0rage,n,i (t+k|t),P pumped,n,j (t+k|t),P VPP,n,l (t+k|t)] T P is the control vector; storage,n,i (t+k|t) represents the power output at time t+k predicted by the i-th energy storage unit at node n based on the time t information; P pumped,n,j (t+k|t) represents the power output at time t+k predicted by the j-th pumped storage unit at node n based on the time t information; P VPP,n,l (t+k|t) represents the aggregated temperature-controlled load at time t+k predicted by the l-th virtual power plant at node n based on the time t information;

[0036] Based on the rolling optimization principle of model predictive control, the above finite-time domain optimization problem is solved in each control cycle to obtain the optimal control sequence in the prediction domain; to ensure feedback correction and real-time adaptability, the control action of the first moment of the optimal control sequence is executed only.

[0037] In the next control cycle t+1, based on the latest system state measurements and load forecast information, the initial conditions and forecast data of the optimization problem are updated, and the optimization problem is solved again.

[0038] Furthermore, in step 5, the objective function of the multi-type energy coordinated frequency regulation centralized optimization model is as follows:

[0039]

[0040] Among them, J cost (t+k|t) represents the instantaneous running cost at time t+k, J smooth (t+k|t) represents the control smoothness penalty term, J terminal (t+N p |t) represents the terminal cost function, and ρ3 and ρ2 are the control smoothness weight and terminal weight, respectively.

[0041] Furthermore, the terminal cost function primarily considers the energy storage SOC state at the end of the prediction domain, ensuring that the energy storage system has the ability to continuously regulate frequency at the end of the prediction domain.

[0042]

[0043] In the formula, I storage,n w represents the total number of energy storage units at node n. terminal,n,i S represents the terminal cost weight of the i-th energy storage unit at node n. target,n,i This represents the target SOC value of the i-th energy storage unit at node n.

[0044] Furthermore, in step 6, the specific content of the Big M method is as follows:

[0045] First, three auxiliary binary variables are introduced to represent the indoor temperature indicator, and the relationship between the auxiliary binary variables and the temperature is established. Then, the three auxiliary binary variables are linked to the operating status of the air conditioning load. Thus, the nonlinear temperature control load operating status constraint is transformed into a mixed integer linear constraint.

[0046] Furthermore, in step 7, to address the computational complexity issue of the centralized optimization model for large-scale multi-type energy coordinated frequency regulation, the alternating direction multiplier method is introduced to decompose the centralized optimization problem into multiple sub-problems, achieving distributed parallel solution. The alternating direction multiplier method decomposes the original problem into local sub-problems and global coordination problems by introducing auxiliary variables and equality constraints. Each subsystem is solved independently, and global optimum is achieved through bivariate updates.

[0047] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0048] 1. This invention establishes a forward-looking optimization framework based on model predictive control. By predicting and optimizing the future state evolution of the system within the prediction domain, it effectively handles complex time-varying characteristics such as changes in the state of charge of energy storage, the thermal inertia evolution of temperature-controlled loads in virtual power plants, and the dynamics of pumped storage capacity. This significantly improves the predictability and system stability of frequency regulation control and overcomes the problem of improper frequency regulation resource allocation caused by traditional passive response modes.

[0049] 2. This invention constructs a comprehensive optimization objective function that includes power generation cost, frequency regulation resource operation cost, energy storage state of charge deviation cost, carbon emission cost, and frequency regulation failure penalty cost. It achieves unified optimization of economy, sustainability, environmental protection, and reliability, and provides important technical support for building a clean, low-carbon, safe, and efficient modern power system.

[0050] 3. This invention transforms the nonlinear constraints of temperature-controlled loads in a virtual power plant into mixed-integer linear constraints using the Big M method, effectively solving the problem of handling complex non-convex constraints, ensuring the solvability of the optimization problem, and enabling centralized optimization models for multi-type energy coordinated frequency regulation to be solved efficiently using commercial solvers.

[0051] 4. This invention uses the alternating direction multiplier method to achieve distributed optimization solutions, decomposing large-scale centralized optimization problems into multiple node subproblems to be solved in parallel, which significantly improves computational efficiency and system scalability. It effectively solves the problems faced by traditional centralized methods, such as exponential growth in computational complexity, excessive communication burden, and risk of single point of failure, and provides a feasible technical path for real-time control of large-scale distributed systems. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the method of the present invention;

[0053] Figure 2 This is a flowchart of the distributed optimization algorithm of the alternating direction multiplier method of the present invention. Detailed Implementation

[0054] The present invention will now be described in further detail with reference to the embodiments and accompanying drawings.

[0055] like Figure 1 As shown, this invention is a distributed optimization control method for coordinated secondary frequency regulation of multiple energy sources considering carbon emissions, comprising the following steps:

[0056] Step 1: Set initial parameters, including technical parameters and operating constraints of the energy storage system, pumped storage, and virtual power plant; power grid topology and line parameters; prediction domain length and control domain length of the model predictive control; collect real-time power demand, energy storage state of charge, temperature-controlled load indoor temperature, and weather forecast data of each node in the power grid through the secondary frequency regulation system involving multiple types of energy.

[0057] Step 2: Establish a mathematical model for the participation of energy storage systems, pumped storage, and virtual power plants in secondary frequency regulation, taking into account the dynamic characteristics, physical constraints, and operational limitations of various frequency regulation resources;

[0058] Step 3: Based on model predictive control, predict the state evolution of the secondary frequency regulation system with multiple types of energy in the prediction domain, and establish a prediction model for the state evolution of energy storage charge and a prediction model for the temperature evolution of temperature-controlled load in a virtual power plant.

[0059] Step 4: Implement a rolling optimization mechanism for the evolution prediction model established in Step 3. In each control cycle, only the control action of the first moment of the optimal control sequence is executed. In the next control cycle, the optimization problem is solved again based on the updated state information of the secondary frequency regulation system with multiple energy types participating, so as to achieve dynamic feedback correction and real-time adaptation.

[0060] Step 5: Construct a centralized optimization model for multi-type energy coordinated frequency regulation that considers carbon emissions, with the goal of minimizing system operating costs, carbon emission costs, and frequency regulation failure penalty costs, while satisfying power balance constraints, generator constraints, and power flow constraints.

[0061] Step 6: The nonlinear operating state constraint of the temperature control load in the virtual power plant is transformed into a mixed integer linear constraint using the Big M method. A linear relationship between the temperature indication variable and the air conditioning operating state is established by introducing auxiliary binary variables.

[0062] Step 7: Based on the alternating direction multiplier method, the centralized optimization problem is decomposed into multiple local optimization problems according to the nodes. Each node solves its local optimization problem independently. Global coordination is achieved through auxiliary variable updates and Lagrange multiplier updates until the optimal solution is converged.

[0063] Mathematical model of energy storage system participating in secondary frequency regulation in step 2

[0064] As a frequency regulation resource with fast response and high regulation accuracy, the dynamic evolution of the State of Charge (SOC) of energy storage systems directly affects their continuous frequency regulation capability. The mathematical model for energy storage systems participating in secondary frequency regulation is established as follows:

[0065] The SOC dynamic evolution equation of the energy storage system is:

[0066]

[0067] Among them, S n,i (t) represents the state of charge of the i-th energy storage unit at node n at time t, P storage,n,i (t) represents the power output of the i-th energy storage unit at node n at time t, E cap,n,i Δt represents the rated capacity of the i-th energy storage unit at node n, and Δt represents the time interval.

[0068] The power constraint of an energy storage system is expressed as:

[0069] P storage,min,n,i ≤P storage,n,i (t)≤P storage,max,n,i

[0070] Among them, P storage,min,n,i and P storage,max,n,i These represent the minimum and maximum power outputs of the i-th energy storage unit at node n, respectively.

[0071] The SOC constraint of an energy storage system is expressed as:

[0072] S min,n,i ≤S n,i (t)≤S max,n,i

[0073] Among them, S min,n,i and S max,n,i Let SOC represent the minimum and maximum values ​​of the i-th energy storage unit at node n, respectively.

[0074] The power change rate constraint of the energy storage system is expressed as:

[0075] |P storage,n,i (t)-P storage,n,i (t-1)|≤R storage,n,i ·Δt

[0076] Among them, R storage,n,i This represents the maximum power change rate of the i-th energy storage unit at node n.

[0077] The aggregate power of the energy storage system at node n is:

[0078]

[0079] Among them, I storage,n This indicates the number of energy storage units configured at node n.

[0080] Mathematical model of pumped storage participating in secondary frequency regulation in step 2

[0081] Pumped storage has the advantages of large capacity and low cost, but its response is relatively slow. The mathematical model for pumped storage participating in frequency regulation is established as follows:

[0082] The power constraint of pumped storage is expressed as:

[0083] -P pumped,max,n,j ≤P pumped,n,j (t)≤P pumped,max,n,j

[0084] Among them, P pumped,n,j (t) represents the power output of the j-th pumped storage unit at node n at time t;

[0085] P pumped,max,n,j This represents the maximum power output of the j-th pumped storage unit at node n at time t. A positive value indicates the power generation mode, and a negative value indicates the pumping mode.

[0086] The upper and lower reservoir capacity constraints of pumped storage are expressed as follows:

[0087] V min,n,j ≤V n,j (t)≤Vmax,n,j

[0088] Among them, V n,j (t) represents the upper reservoir water level of the j-th pumped storage unit at node n at time t, V min,n,j and V max,n,j Let represent the minimum and maximum upper reservoir water levels of the j-th pumped storage unit at node n at time t, respectively.

[0089] The power change rate constraint for pumped storage is expressed as:

[0090] |P pumped,n,j (t)-P pumped,n,j (t-1)|≤R pumped,n,j ·Δt

[0091] Among them, R pumped,n,j This represents the maximum power change rate of the j-th pumped storage unit at node n.

[0092] The pumped storage power aggregate of node n is:

[0093]

[0094] Among them, I pumped,n This indicates the number of pumped storage units configured at node n.

[0095] Mathematical model of virtual power plant participating in secondary frequency regulation in step 2

[0096] Virtual power plants aggregate distributed resources to form frequency regulation capabilities, featuring abundant resources and moderate costs. Taking temperature-controlled loads as an example, a mathematical model is established for distributed load resources to participate in frequency regulation in a virtual power plant, utilizing the thermal inertia of temperature-controlled loads to participate in auxiliary frequency regulation.

[0097] The temperature control load is modeled using a first-order equivalent thermal parameter model to describe the temperature change in the air-conditioned room. The first-order differential equation of the equivalent thermal parameter model is:

[0098]

[0099] Where C represents the equivalent heat capacity, R represents the equivalent thermal resistance, and T in (t) and T out Q(t) represents the indoor and outdoor ambient temperatures at time t, respectively; Q(t) represents the cooling / heating output of the air conditioner at time t. When Q(t) < 0, the air conditioner is in cooling mode; when Q(t) > 0, the air conditioner is in heating mode.

[0100] Discretizing the above first-order differential equation yields the first-order recursive formula for indoor ambient temperature:

[0101]

[0102] Among them, T in,n,k,j (t) represents the indoor ambient temperature of the j-th air conditioner in the k-th virtual power plant at node n at time t. out,n,k,j (t) represents the outdoor ambient temperature (C) of the j-th air conditioner in the k-th virtual power plant at node n at time t. n,k,j R represents the equivalent heat capacity of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j Let Q represent the equivalent thermal resistance of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j (t) represents the cooling / heating of the j-th air conditioner in the k-th virtual power plant at node n at time t.

[0103] The cooling / heating constraint of an air conditioner is expressed as:

[0104] Q n,k,j (t)=η n,k,j ·P n,k,j (t)·S n,k,j (t)

[0105] 0≤P n,k,j (t)≤P rated,n,k,j

[0106] Where, η n,k,j P represents the energy efficiency conversion coefficient of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j (t) represents the power of the j-th air conditioner in the k-th virtual power plant at node n at time t, S n,k,j (t) represents the working state of the j-th air conditioner in the k-th virtual power plant at node n at time t (1 for running, 0 for off), P rated,n,k,j This represents the rated power of the j-th air conditioner in the k-th virtual power plant at node n.

[0107] Assume the user sets the temperature value to T. set,n,k,j The air conditioner will then adjust its power to maintain the indoor temperature within the permissible range.

[0108] T min,n,k,j ≤T in,n,k,j (t)≤T max,n,k,j

[0109] T min,n,k,j =T set,n,k,j -ΔT n,k,j

[0110] T max,n,k,j =T set,n,k,j +ΔT n,k,j

[0111] Among them, T min,n,k,j and T max,n,k,jLet ΔT represent the minimum and maximum allowable indoor temperatures of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j This represents the allowable temperature deviation of the j-th air conditioner in the k-th virtual power plant at node n.

[0112] Taking air conditioning cooling mode as an example, when the air conditioner is off, the indoor temperature gradually rises; when the air conditioner is running, the indoor temperature gradually decreases. The air conditioner maintains a temperature of T... min,n,k,j and T max,n,k,j Within the specified interval, the on / off state is continuously switched. Taking the cooling state as an example, the operating state constraint of the air conditioner is expressed as:

[0113]

[0114] The aggregated temperature-controlled load in the k-th virtual power plant at node n is:

[0115]

[0116] Among them, Ω n,k This represents the set of temperature-controlled loads in the k-th virtual power plant at node n.

[0117] Therefore, the up-frequency modulation capacity P of the temperature-controlled load participating in the system frequency modulation can be obtained. VPP,up,n,k (t) and down-modulation capacity P VPP,down,n,k (t):

[0118]

[0119] The aggregated power of the virtual power plant at node n is:

[0120]

[0121] Among them, I VPP,n This indicates the number of virtual power plants configured in node n.

[0122] The multi-step prediction framework based on model predictive control (hereinafter referred to as the model predictive control framework) formed by steps 3 and 4.

[0123] To improve the foresight and system stability of frequency modulation control, model predictive control (MPC) is introduced to predict and optimize the future state of the system within the prediction domain.

[0124] Define the prediction domain length as N p The control domain length is N c , where N c ≤N p At the current time t, the model predictive controller needs to predict the time from t+1 to t+N. p The system state evolution is analyzed, and optimization is performed from time t to t+N. c-1 control sequence.

[0125] The SOC evolution prediction model (i.e., the energy storage state of charge evolution prediction model) of the energy storage system in the prediction domain is expressed as follows:

[0126]

[0127] Where k = 1, 2, ..., N p S n,i (t+k|t) represents the SOC value at time t+k predicted by the i-th energy storage unit at node n based on the time t information, P storage,n,i (t+k-1|t) represents the power output of the i-th energy storage unit at node n predicted at time t+k-1 based on the time t information.

[0128] The temperature evolution prediction model of temperature-controlled load in the prediction domain in a virtual power plant (i.e., the temperature evolution prediction model of temperature-controlled load in a virtual power plant) is expressed as follows:

[0129]

[0130] Among them, T in,n,k,j (t+k-1|k) represents the indoor temperature at time t+k-1 predicted by the j-th air conditioner in the t-th virtual power plant at node n based on the time information t; Δt represents the time interval; T out,n,k,j (t+k-1|t) represents the outdoor temperature at time t+k-1 predicted by the j-th air conditioner in the k-th virtual power plant at node n based on time t information, which can be obtained based on weather forecast data; C n,k,j R represents the equivalent heat capacity of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j Let Q represent the equivalent thermal resistance of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j (t+k-1|t) represents the cooling / heating capacity of the j-th air conditioner in the k-th virtual power plant at node n, predicted at time t+k based on the time t information.

[0131] To account for prediction uncertainties, a rolling optimization mechanism is introduced within the prediction domain. In each control cycle, the following finite-time optimization problem is solved:

[0132]

[0133] Where J(t+k|t) represents the instantaneous cost at time t+k, ΔU(t+k|t) represents the control increment, ρ1 is the control increment weight, used to suppress excessively frequent control actions; J terminal (t+N p |t) is the terminal cost function, and ρ2 is the terminal weight.

[0134] The aforementioned real-time operating costs include power generation costs, frequency regulation resource operating costs, energy storage SOC deviation costs, carbon emission costs, and frequency regulation failure penalty costs.

[0135] The control increment is defined as:

[0136] ΔU(t+k|t)=U(t+k|t)-U(t+k-1|t)

[0137] Where, U(t+k|t)=[P storage,n,i (t+k|t),P pumped,n,j (t+k|t),P VPP,n,l (t+k|t)] T P is the control vector. storage,n,i (t+k|t) represents the power output at time t+k predicted by the i-th energy storage unit at node n based on the time t information; P pumped,n,j (t+k|t) represents the power output at time t+k predicted by the j-th pumped storage unit at node n based on the time t information; P VPP,n,l (t+k|t) represents the aggregated temperature-controlled load at time t+k predicted by the l-th virtual power plant at node n based on the time t information.

[0138] Under the rolling optimization strategy (mechanism), only the first element of the optimal control sequence is executed:

[0139] U (t) =U (t|t)

[0140] Then, at the next time step t+1, the optimization problem is solved again based on the updated system state information to achieve feedback correction.

[0141] Step 5) Multi-type energy coordinated frequency regulation centralized optimization model

[0142] Based on the established model predictive control framework, a centralized optimization model for coordinated frequency regulation of multiple energy types within the prediction domain is constructed. This model fully utilizes the forward-looking optimization capability of MPC, comprehensively considering system state evolution, multi-objective coordination, and constraint satisfaction within the prediction domain, to achieve dynamic and unified optimization of economy, sustainability, environmental protection, and reliability.

[0143] At each time point within the prediction domain, the system needs to satisfy the active and reactive power balance constraints:

[0144] ∑ h∈N P nh (t+k|t)=P G,n (t+k|t)-P load,base,n (t+k|t)-ΔP command,n (t+k|t)+P fail,n (t+k|t)

[0145] Where k = 0, 1, 2, ..., N p -1 represents the time step within the prediction domain, N represents the set of nodes in the power grid, and P nh (t+k|t) represents the active power flowing from node n to node h at time t+k, predicted based on time t information. G,n (t+k|t) represents the active power of generator n at node t+k predicted based on information at time t; P load,base,n (t+k|t) represents the load predicted at time t+k based on the information at time t; ΔP command,n (t+k|t) represents the command to calculate the frequency regulation power change of each energy source at time t+k based on the information at time t; P fail,n (t+k|t) represents the frequency modulation failure power calculated based on the information at time t+k.

[0146] Frequency modulation power change commands are handled collaboratively by various frequency modulation resources within the prediction domain:

[0147] ΔP command,n (t+k|t)=P storage,n (t+k|t)+P pumped,n (t+k|t)+P VPP,n (t+k|t)

[0148] Generators within the prediction domain must satisfy power constraints and ramp rate constraints:

[0149] P G,min,n ≤P G,n (t+k|t)≤P G,max,n

[0150] Q G,min,n ≤Q G,n (t+k|t)≤Q G,max,n

[0151] P G,n (t+k|t)-P G,n (t+k-1|t)≤R up,n ·Δt

[0152] P G,n (t+k-1|t)-P G,n (t+k|t)≤R down,n ·Δt

[0153] Among them, P G,min,n and P G,max,n Let Q represent the maximum and minimum active power of the generator at node n, respectively. G,min,n and Q G,max,n R represents the maximum and minimum reactive power of the generator at node n, respectively.up,n and R down,n These represent the maximum upward and downward ramp rates of the generator at node n, respectively.

[0154] The energy storage state of charge evolution prediction model established in step 3 and the temperature evolution prediction model of temperature-controlled load in the virtual power plant are integrated as optimization constraints to ensure that the dynamic evolution of energy storage SOC and temperature-controlled load temperature is reasonably controlled.

[0155] Construct an optimization objective function for a multi-energy coordinated frequency regulation centralized optimization model within the prediction domain, comprehensively considering the system's economic efficiency, environmental friendliness, reliability, and control smoothness:

[0156]

[0157] Among them, J cost (t+k|t) represents the instantaneous running cost at time t+k, J smooth (t+k|t) represents the control smoothness penalty term, J terminal (t+N p |t) represents the terminal cost function, and ρ3 and ρ2 are the control smoothness weight and terminal weight, respectively.

[0158] The components of real-time operating costs are represented as follows:

[0159]

[0160] C operation,n (t+k|t)=C storage,op,n (t+k|t)+C pumped,op,n (t+k|t)+C VPP,op,n (t+k|t),

[0161]

[0162] C carbon,n (t+k|t)=λ carbon μ n ·P G,n (t+k|t);

[0163]

[0164] Among them, C generation,n C represents the power generation cost of the generator at node n. operation,n C represents the operating cost of multiple energy sources at node n. storage,op,n C pumped,op,n C VPP,op,n C represents the operating costs of the energy storage unit, pumped storage unit, and virtual power plant at node n, respectively. carbon,n C represents the carbon emission cost of the generator at node n.penalty,n a represents the penalty cost for frequency modulation failure at node n. n b represents the coefficient of the quadratic term of the generator power generation cost at node n. n a represents the coefficient of the first-order term of the generator power generation cost at node n. storage,n,i b represents the coefficient of the quadratic term of the operating cost of the i-th energy storage unit at node n. storage,n,i a represents the coefficient of the first-order term of the operating cost of the i-th energy storage unit at node n. pumped,n,j b represents the coefficient of the quadratic term of the operating cost of the i-th pumped storage unit at node n. pumped,n,j a represents the coefficient of the linear term of the operating cost of the j-th pumped storage unit at node n. VPP,n,l b represents the coefficient of the quadratic term of the operating cost of the l-th virtual power plant at node n. VPP,n,l λ represents the coefficient of the linear term of the operating cost of the l-th virtual power plant at node n. carbon λ represents the unit carbon emission cost coefficient. fail μ represents the penalty cost coefficient for frequency modulation failure. n This represents the carbon emission coefficient per unit of electricity generated by the generator at node n.

[0165] To avoid drastic changes in frequency modulation commands, a control smoothness penalty term is introduced:

[0166]

[0167] in, These represent the control smoothness weights for various frequency modulation resources.

[0168] The terminal cost function primarily considers the energy storage SOC state at the end of the prediction domain, ensuring that the energy storage system has continuous frequency regulation capability at the end of the prediction domain.

[0169]

[0170] In the formula, I storage,n w represents the total number of energy storage units at node n. terninal,n,i S represents the terminal cost weight of the i-th energy storage unit at node n. target,n,i This represents the target SOC value of the i-th energy storage unit at node n.

[0171] Based on the rolling optimization principle of MPC, the above finite-time optimization problem is solved in each control cycle to obtain the optimal control sequence in the prediction domain. To ensure feedback correction and real-time adaptability, the control action of the first moment of the optimal control sequence is executed only:

[0172]

[0173] In the formula, the "*" in the upper right corner of each symbol indicates the predicted value of that symbol.

[0174] In the next control cycle t+1, based on the latest system state measurements and load forecast information, update the initial conditions and forecast data of the optimization problem, and resolve the optimization problem:

[0175]

[0176] In the formula, This represents the SOC measurement value of the i-th energy storage unit at node n. This represents the measured indoor temperature of the j-th air conditioner in the k-th virtual power plant at node n.

[0177] This rolling optimization mechanism ensures that the control strategy can respond promptly to changes in system state and external disturbances, achieving dynamic, coordinated, and optimized control of multiple types of energy sources.

[0178] In step 6), the nonlinear constraint of the temperature control load is transformed into a mixed-integer linear constraint using the Big M method.

[0179] Since the working state constraints of the temperature-controlled load in the virtual power plant are non-convex, it is difficult to solve the temperature evolution prediction model of the temperature-controlled load directly using a solver. Therefore, the non-convex constraints are transformed into mixed integer linear constraints by using the Big M method.

[0180] First, we introduce three auxiliary binary variables α. n,k,j (t), β n,k,j (t), γ n,k,j Let (t) represent the indicator variable for indoor temperature. The relationship between the auxiliary binary variable and temperature is established as follows:

[0181] T in,n,k,j (t)-T max,n,k,j ≤M·(1-α n,k,j (t))

[0182] T min,n,k,j -T in,n,k,j (t)≤M·(1-β n,k,j (t))

[0183] T max,n,k,j -T in,n,k,j (t)≤M·γ n,k,j (t)

[0184] T in,n,k,j (t)-T min,n,k,j ≤M·γ n,k,j (t)

[0185] Here, M represents a very large number used to relax the constraints.

[0186] Then, the operating status S of the three auxiliary binary variables and the air conditioning load will be used. n,k,j(t) Establish the following relationship:

[0187] S n,k,j (t)≥α n,k,j (t)

[0188] S n,k,j (t)≤1-β n,k,j (t)

[0189] S n,k,j (t)-S n,k,j (t-1)≤α n,k,j (t)

[0190] S n,k,j (t-1)-S n,k,j (t)≤β n,k,j (t)

[0191] α n,k,j (t)+β n,k,j (t)+γ n,k,j (t)=1

[0192] Therefore, the nonlinear operating state constraint of the temperature-controlled load is transformed into a mixed integer linear constraint, so that the temperature evolution prediction model of the temperature-controlled load can be directly solved using a commercial solver.

[0193] In step 7), the alternating direction multiplier method (ADMM) is used to achieve distributed optimization solution.

[0194] To address the computational complexity issue of a centralized optimization model for large-scale, multi-type energy coordinated frequency regulation (i.e., a centralized optimization problem), the Alternating Direction Multiplier Method (ADMM) is introduced to decompose the centralized optimization problem into multiple subproblems, enabling distributed parallel solution. The ADMM algorithm decomposes the original problem into local subproblems and a global coordination problem by introducing auxiliary variables and equality constraints. Each subsystem can be solved independently, and global optimality is achieved through bivariate updates.

[0195] The centralized optimization problem established in step 5 is decomposed according to nodes. The original optimization problem can be represented as:

[0196]

[0197] Among them, f n (x n ) represents the local objective function of node n, which includes the node's power generation cost, frequency regulation resource operation cost, energy storage SOC deviation cost, carbon emission cost, etc.; g nh (x n ,x h ) represents the coupling term between nodes n and h, mainly reflected in power flow constraints; x n The decision variables for node n include P.G,n P storage,n,i P pumped,n,j P VPP,n,l etc.; χ n Let ε represent the set of local constraints for node n; let ε represent the set of paths in the network.

[0198] To apply the ADMM algorithm, an auxiliary variable z is introduced. nh Represent the power flow on line nh and establish equality constraints:

[0199]

[0200] Based on the ADMM framework, construct the augmented Lagrangian function:

[0201]

[0202] Where, λ nh ρ represents the Lagrange multiplier corresponding to the equality constraint, and ρ>0 represents the penalty parameter.

[0203] By using ADMM to solve system optimization problems in a distributed manner, such as Figure 2 As shown, the power system is divided into multiple regions based on geographical location or electrical connection characteristics, and a system dispatch center is established for global coordination. Each region defines an auxiliary power variable z on its boundary lines. nh And apply consistency constraints P through ADMM nh (x n ,x h )=z nh The line power z between the two regions nh They exchange information as coordinating variables.

[0204] The ADMM algorithm achieves distributed solution by alternately updating three sets of variables. Each node n independently solves the following local optimization subproblem:

[0205]

[0206] This subproblem includes all local constraints of node n, including generator power constraints and ramp rate constraints, dynamic evolution constraints, power constraints, and rate of change constraints of the energy storage system's SOC, power constraints, reservoir capacity constraints, and rate of change constraints of pumped storage, dynamic temperature constraints and operating state constraints of the virtual power plant's temperature-controlled load (linearized using the Big M method), and node power balance constraints.

[0207] Update auxiliary variable z nh The problem has an analytical solution:

[0208]

[0209] in, Based on the updated node variables and Calculated.

[0210] Update the Lagrange multipliers:

[0211]

[0212] By using ADMM decomposition, we can obtain the optimization subproblems for each sub-region. Taking a two-region system as an example, the optimization subproblem for sub-region one is expressed as:

[0213]

[0214] In the formula, J cost,1 J represents the real-time operating cost of a sub-region at any given moment; smooth,1 J represents the smoothness penalty term for each time step in the sub-region; terminal,1 This represents the terminal cost function at each time step in a sub-region.

[0215] The sub-region binary optimization subproblem is represented as:

[0216]

[0217] Where λ1 and λ2 are the Lagrange multipliers of each region's sub-model, respectively, and P nh Let z be the exchange power between region n and its neighbor h. nh J is the corresponding auxiliary variable. cost,2 J represents the real-time operating cost of sub-region two at each moment; smooth,2 J represents the smoothness penalty term for each time step in the sub-region; terminal,2 This represents the smoothness penalty term for each time step in sub-region two.

[0218] To improve the convergence speed of the algorithm, an accelerated ADMM method with relaxation parameters β∈(0,2) is introduced. When β>1, it is called over-relaxation. In distributed optimization solutions, considering the results of the previous iteration, relaxation techniques are used to update variables more effectively. The modified algorithm update formula is as follows:

[0219]

[0220] λ k =λ k-1 +(x k -z k )

[0221] In the formula, z k Let x represent the auxiliary variable for the k-th iteration. k Let λ represent the decision variable for the k-th iteration. k Let k represent the Lagrange multiplier in the k-th iteration. gG(g) represents the number of coupled variables connected to the global variable, and G(g) represents the mapping relationship between the global variable and each region.

[0222] Define the original residual and the dual residual to determine the convergence of the algorithm:

[0223]

[0224] When the convergence condition |r is satisfied (k+1) |2≤∈ pri and |s (k+1) |2≤∈ dual When the algorithm terminates, ∈ pri and ∈ dual These represent the tolerances for the original and dual residuals, respectively.

[0225] The ADMM distributed optimization solution framework described above can decompose the large-scale multi-type energy coordinated frequency regulation optimization problem that originally required centralized processing into multiple parallel sub-problems, which significantly reduces computational complexity, improves solution efficiency, and at the same time ensures the privacy of information in each region, thus realizing distributed coordinated optimization control of multi-type energy resources.

[0226] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A distributed optimization control method for multi-type energy collaborative secondary frequency modulation considering carbon emissions, characterized in that, include: Step 1: Set initial parameters, including technical parameters and operating constraints of the energy storage system, pumped storage, and virtual power plant; power grid topology and line parameters; prediction domain length and control domain length of the model predictive control; collect real-time power demand, energy storage state of charge, temperature-controlled load indoor temperature, and weather forecast data of each node in the power grid through the secondary frequency regulation system involving multiple types of energy. Step 2: Establish a mathematical model for the participation of energy storage systems, pumped storage, and virtual power plants in secondary frequency regulation, taking into account the dynamic characteristics, physical constraints, and operational limitations of various frequency regulation resources; Step 3: Based on model predictive control, predict the state evolution of the secondary frequency regulation system with multiple types of energy in the prediction domain, and establish a prediction model for the state evolution of energy storage charge and a prediction model for the temperature evolution of temperature-controlled load in a virtual power plant. Step 4: Implement a rolling optimization mechanism for the evolution prediction model established in Step 3. In each control cycle, only the control action of the first moment of the optimal control sequence is executed. In the next control cycle, the optimization problem is solved again based on the updated state information of the secondary frequency regulation system with multiple energy types participating, so as to achieve dynamic feedback correction and real-time adaptation. Step 5: Construct a centralized optimization model for multi-type energy coordinated frequency regulation that considers carbon emissions, with the goal of minimizing system operating costs, carbon emission costs, and frequency regulation failure penalty costs, while satisfying power balance constraints, generator constraints, and power flow constraints. Step 6: The nonlinear operating state constraint of the temperature control load in the virtual power plant is transformed into a mixed integer linear constraint using the Big M method. A linear relationship between the temperature indication variable and the air conditioning operating state is established by introducing auxiliary binary variables. Step 7: Based on the alternating direction multiplier method, the centralized optimization problem is decomposed into multiple local optimization problems according to the nodes. Each node solves its local optimization problem independently. Global coordination is achieved through auxiliary variable updates and Lagrange multiplier updates until the optimal solution is converged.

2. The multi-type energy coordinated secondary frequency modulation distributed optimization control method according to claim 1, characterized in that, In step 2), A mathematical model for energy storage systems participating in secondary frequency regulation is established, including the dynamic evolution equation of the energy storage system's state of charge (SOC), power constraints, SOC constraints, and power change rate constraints. A mathematical model for pumped storage participating in secondary frequency regulation is established, including power constraints, upper and lower reservoir capacity constraints, and power change rate constraints for pumped storage. A mathematical model for virtual power plants participating in secondary frequency regulation is established, including temperature dynamic change equations and operating state constraints.

3. The multi-type energy collaborative secondary frequency modulation distributed optimization control method according to claim 1, characterized in that, The aforementioned model predictive control defines a prediction domain length of N. p The control domain length is N c N c ≤N p At the current time t, the model predictive controller needs to predict the time from time t+1 to t+N. p Multiple energy sources participate in the state evolution of the secondary frequency regulation system, and the process from time t to t+N is optimized. c -1 control sequence.

4. The multi-type energy coordinated secondary frequency modulation distributed optimization control method according to claim 3, characterized in that, In step 3, the energy storage charge state evolution prediction model is expressed as follows: Where k = 1, 2, ..., N p S n,i (t+k|t) represents the SOC value at time t+k predicted by the i-th energy storage unit at node n based on the time t information, P storage,n,i (t+k-1|t) represents the power output at time t+k-1 predicted by the i-th energy storage unit at node n based on the time t information, E cap,n,i Δt represents the rated capacity of the i-th energy storage unit at node n, and Δt represents the time interval.

5. The multi-type energy coordinated secondary frequency modulation distributed optimization control method according to claim 3, characterized in that, In step 3, the temperature evolution prediction model for temperature-controlled loads in the virtual power plant is expressed as follows: Among them, T in,n,k,j (t+k-1|t) represents the indoor temperature at time t+k-1 predicted by the j-th air conditioner in the k-th virtual power plant at node n based on the time t information, where Δt represents the time interval T. out,n,k,j (t+k-1|t) represents the outdoor temperature at time t+k-1 predicted by the j-th air conditioner in the k-th virtual power plant at node n based on the time t information, C n,k,j R represents the equivalent heat capacity of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j Let Q represent the equivalent thermal resistance of the j-th air conditioner in the k-th virtual power plant at node n. n,k,j (t+t-1|t) represents the cooling / heating capacity of the j-th air conditioner in the t-th virtual power plant at node n, predicted at time t+k based on the time t information.

6. The multi-type energy collaborative secondary frequency modulation distributed optimization control method according to claim 3, characterized in that, In step 4, to account for prediction uncertainty, a rolling optimization mechanism is introduced within the prediction domain. In each control cycle, the following finite-time optimization problem is solved: Where J(t+k|t) represents the instantaneous cost at time t+k, ΔU(t+k|t) represents the control increment, ρ1 is the control increment weight, and J terminal (t+N p |t) is the terminal cost function, and ρ2 is the terminal weight; the real-time operating cost includes power generation cost, frequency regulation resource operating cost, energy storage SOC deviation cost, carbon emission cost, and frequency regulation failure penalty cost; The control increment is defined as: ΔU(t+k|t)=U(t+k|t)-U(t+k-1|t) Where, U(t+k|t)=[P storage,n,i (t+k|t),P pumped,n,j (t+k|t),P VPP,n,l (t+k|t)] T P is the control vector; storage,n,i (t+k|t) represents the power output at time t+k predicted by the i-th energy storage unit at node n based on the time t information; P pumped,n,j (t+k|t) represents the power output at time t+k predicted by the j-th pumped storage unit at node n based on the time t information; P VPP,n,l (t+k|t) represents the aggregated temperature-controlled load at time t+k predicted by the l-th virtual power plant at node n based on the time t information; Based on the rolling optimization principle of model predictive control, the above finite-time domain optimization problem is solved in each control cycle to obtain the optimal control sequence in the prediction domain; to ensure feedback correction and real-time adaptability, the control action of the first moment of the optimal control sequence is executed only. In the next control cycle t+1, based on the latest system state measurements and load forecast information, the initial conditions and forecast data of the optimization problem are updated, and the optimization problem is solved again.

7. The multi-type energy collaborative secondary frequency modulation distributed optimization control method according to claim 6, characterized in that, In step 5, the objective function of the multi-energy coordinated frequency regulation centralized optimization model is as follows: Among them, J cost (t+k|t) represents the instantaneous running cost at time t+k, J smooth (t+k|t) represents the control smoothness penalty term, J terminal (t+N p |t) represents the terminal cost function, and ρ3 and ρ2 are the control smoothness weight and terminal weight, respectively.

8. The multi-type energy collaborative secondary frequency regulation distributed optimization control method according to claim 7, characterized in that, The terminal cost function mainly considers the energy storage SOC state at the end of the prediction domain, ensuring that the energy storage system has the ability to continuously regulate frequency at the end of the prediction domain. In the formula, I storage,n w represents the total number of energy storage units at node n. terminal,n,i S represents the terminal cost weight of the i-th energy storage unit at node n. target,n,i This represents the target SOC value of the i-th energy storage unit at node n.

9. The multi-type energy collaborative secondary frequency regulation distributed optimization control method according to claim 7, characterized in that, In step 6, the specific details of the Big M method are as follows: First, three auxiliary binary variables are introduced to represent the indoor temperature indicator, and the relationship between the auxiliary binary variables and the temperature is established. Then, the three auxiliary binary variables are linked to the operating status of the air conditioning load, thereby transforming the nonlinear temperature control load operating status constraint into a mixed integer linear constraint.

10. The multi-type energy collaborative secondary frequency modulation distributed optimization control method according to claim 7, characterized in that, In step 7, to address the computational complexity of the centralized optimization model for large-scale multi-type energy coordinated frequency regulation, the alternating direction multiplier method is introduced to decompose the centralized optimization problem into multiple sub-problems, enabling distributed parallel solution. The alternating direction multiplier method decomposes the original problem into local sub-problems and global coordination problems by introducing auxiliary variables and equality constraints. Each subsystem is solved independently, and global optimum is achieved through bivariate updates.

Images