Regional energy internet hierarchical control method considering flexible load

Abstract

The invention discloses a regional energy internet hierarchical control method considering flexible loads, which comprises the following steps of: constructing a global layer, dividing a regional energy internet into a plurality of electrically strongly coupled sub-regions by adopting a dynamic partitioning algorithm, aggregating equivalent adjustable capacities of the flexible loads in the sub-regions, and performing rolling optimization in an hour-level time scale, an improved Benders decomposition algorithm is adopted to accelerate solving and generate a partition-level power regulation instruction; constructing a regional layer, designing a thermodynamic-economic hybrid constraint model according to flexible load physical characteristics and user behavior elasticity, introducing a user comfort recovery time constraint and elastic electricity price response function, and decomposing an adjustment instruction issued by the global layer to each flexible load cluster in a minute-level time scale by adopting a parallel ADMM optimization algorithm; and constructing a load layer, constructing a millisecond-level adjustable load priority switching mechanism based on FPGA hardware acceleration and finite-state machine logic, calculating a load adjustable power boundary in real time, and dynamically adjusting a response queue in combination with system frequency deviation.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a hierarchical and graded control method for a regional energy internet taking flexible loads into consideration. Background Art

[0002] With the penetration rate of renewable energy exceeding 30% and the large-scale access of distributed energy (photovoltaic, energy storage, electric vehicles, etc.), the regional energy Internet presents the characteristics of deep coupling of "source-grid-load-storage". Flexible loads (such as temperature control equipment, interruptible industrial loads, and electric vehicle clusters) are the core regulation resources on the demand side, and their dynamic response capabilities directly affect the supply and demand balance and operational efficiency of the system. However, existing hierarchical control technologies generally adopt a rigid architecture of "centralized optimization + local control". The upper layer performs global economic scheduling on an hourly time scale, and the lower layer relies on fixed logic minute-level equipment adjustment. It is difficult to adapt to the coordinated needs of high-proportion renewable energy volatility and flexible load second-level response, resulting in limited system regulation capabilities and low resource utilization.

[0003] The defects of existing technologies can be attributed to the following causal chain: First, the large difference between the upper and lower control cycles (hourly and minute-level) leads to a lack of dynamic coordination mechanism across time scales in the event of sudden disturbances (such as a sudden drop in wind and solar power output), causing overshoot of more than 15% and an increase of 5-8% in wind and solar power curtailment rates; second, the flexible load model is overly simplified (using a 0 / 1 switch model), ignoring its thermodynamic inertia and user behavior elasticity, resulting in only 40-60% of the actual adjustable capacity being utilized, and more than 30% of the demand-side regulation potential being wasted; finally, the variable dimension of the centralized optimization model is O(n 2 ) increases, causing the solution time for thousand-node systems to exceed 10 minutes, unable to meet the real-time control requirements within 5 minutes, resulting in delayed optimization results. These defects collectively lead to a significant decline in system economy, flexibility, and computational efficiency. Summary of the Invention

[0004] The purpose of the present invention is to propose a hierarchical and graded control method for regional energy internet taking flexible loads into account, so as to improve the technical problem that the existing hierarchical control technology is difficult to adapt to the coordinated requirements of high-proportion renewable energy volatility and flexible load second-level response, resulting in limited system regulation capacity and low resource utilization.

[0005] To solve the above technical problems, the present invention provides a hierarchical control method for a regional energy internet taking flexible loads into account, comprising the following steps:

[0006] S1: Construct a global layer. Using electrical coupling analysis and prediction technology, a dynamic partitioning algorithm based on energy transmission margin is designed to divide the regional energy internet into multiple sub-regions with strong electrical coupling. The equivalent adjustable capacity of the flexible loads in each sub-region is aggregated. The global layer is optimized on an hourly timescale, and an improved Benders decomposition algorithm is used to accelerate the solution and generate partition-level power regulation instructions.

[0007] S2: Build a regional layer and design a thermal-economic hybrid constraint model based on the physical characteristics of flexible loads and the elasticity of user behavior. This model introduces user comfort recovery time constraints and a flexible electricity price response function. Using a parallel ADMM optimization algorithm, the regional power regulation instructions issued by the global layer are decomposed into each flexible load cluster within a minute-level time scale.

[0008] S3 builds the load layer. Based on FPGA hardware acceleration and finite state machine logic, it constructs a millisecond-level adjustable load priority switching mechanism, calculates the load adjustable power boundary in real time, and dynamically adjusts the response queue based on the system frequency deviation.

[0009] Preferably, the step S1 specifically includes:

[0010] S11, data preprocessing and electrical coupling calculation, calculate the electrical coupling EC between all adjacent nodes ij The expression is:

[0011]

[0012] Where: P ij (t) is the real-time line transmission power, is the upper limit of line capacity, EC ij The closer it is to 1, the closer the transmission power of line ij is to the limit, and the stronger the coupling between nodes i and j is;

[0013] For each node i, generate a set of strongly coupled neighbors: Nb(i) = {j|EC ij ≥γ}where: γ is the threshold;

[0014] S12, a dynamic partitioning algorithm is used to merge dynamic partitions, merging strongly coupled nodes into the same control area to reduce cross-area power interaction;

[0015] S13, aggregate partition adjustable capacity;

[0016] In S14, the improved Benders decomposition algorithm is used to accelerate the solution and generate the adjustment instructions, wherein the objective function is set to minimize the total system cost:

[0017] Where: Each partition adjustment amount ΔP k is the decision variable, Adjust instructions for each partition, is the cost of power generation, Punishment for abandonment. Forecast output of renewable energy in zone k, Contribute to actual renewable energy, is the regulation instruction of the previous period, γ is the penalty coefficient for wind and solar power abandonment, It is a smoothing term to avoid drastic fluctuations in instructions;

[0018] The constraints include power balance constraint and adjustable capacity limit, which are expressed as:

[0019] Where: ΔP sys is the total system shortfall;

[0020] S15, issue instructions and correction feedback, and output the target adjustment amount for each partition (k=1,2,...,k), set the dynamic correction mechanism: the feedback signal is the actual adjustment amount of the regional layer The modified rule is that if there are 3 consecutive cycles Repartition.

[0021] Wherein, the step S12 is specifically as follows:

[0022] S121, perform initial partitioning, each node forms a partition Ω k ={k},(k=1,2,...,N);

[0023] S122, iterative merging rules:

[0024] Condition: If partition Ω a and Ω b There are common strongly coupled nodes And j∈Nb(i), then merge into a new block Ω a+b ;

[0025] Termination condition: There are no more partitions to merge;

[0026] S123, partition validity check, each partition must meet the lower limit of the total internal coupling degree:

[0027]

[0028] Where: θ is the density threshold;

[0029] If it is not satisfied, the partition is split into smaller sub-areas;

[0030] S124, output the final partition set {Ω1,Ω2,...,Ω K};

[0031] Wherein, the step S13 is specifically as follows:

[0032] S131, input load data, including: each partition node set Ω k , flexible load parameter C i , temperature dead zone Response time User engagement α i (t);

[0033] S132, calculate the adjustable power of a single load,

[0034] For temperature-controlled loads, the adjustable power is the maximum power that can be adjusted per unit time within the allowable temperature deviation range. The expression is:

[0035] For electric vehicles, the expression is:

[0036] Where: Charging willingness coefficient for users, SOC i (t) is the state of charge of the electric vehicle, For electric

[0037] The rated charge and discharge power of the car;

[0038] S133, dynamically adjust user participation, the model expression is:

[0039]

[0040] Where: is the user's basic participation, which is determined by the user contract, ρ(t) is the real-time electricity price, ρ base is the benchmark electricity price, k p is the electricity price elasticity coefficient, For temperature deviation, when the actual temperature is close to the dead zone boundary, the reference

[0041] With the degree of reduction;

[0042] S134, partition capacity aggregation calculation, the expression is:

[0043] Where: β k is the partition weight coefficient, which is dynamically adjusted according to the importance of the partition;

[0044] The dynamic weight adjustment rule is: if the completion rate of the adjustment instruction of partition k in the past hour is less than 90%, then reduce its weight, the expression is: β k =max(0.5,β k -0.1·(1-completion rate));

[0045] S135: Perform capacity verification, including physical constraint verification and economic verification. The physical constraint verification ensures that the aggregation result does not exceed the limit of the zoned power grid equipment:

[0046] Where: is the aggregate capacity, For critical line transmission capacity, is the transformer capacity;

[0047] Economic Verification: If the partition adjustment cost is higher than the threshold, the high-cost partition is marked and its adjustment amount is preferentially reduced in the global optimization.

[0048] Preferably, the improved Benders decomposition algorithm is used to accelerate the solution and generate the adjustment instructions as follows:

[0049] S141, initializing parameters, setting the initial power allocation value of partition k to: k=1,2,...,K

[0050] Where: ΔP total is the total power adjustment;

[0051] S142, the objective function of the main problem is F1, and the relaxed integer variable constraint ΔP is introduced. k =x k , output continuous partition power allocation value;

[0052] S143, the sub-problem is set to verify the feasibility of each partition constraint, generate a cutting plane and return to the main problem. The objective function is:

[0053]

[0054] Where: ΔP k is the power regulation of partition k, ΔP 1,k The power adjustment amount of partition k outputted by the main problem;

[0055] Verify the partition power adjustment ΔP k Whether the coupling constraint and user comfort constraint are satisfied. The coupling constraint is the mutual restriction relationship between the power regulation quantities of each partition, and the expression is: (i=1,2,...,M)

[0056] Where: A ik is the sensitivity coefficient of partition k to constraint i, b i is the upper capacity limit of constraint i;

[0057] The user comfort constraint expression is:

[0058] Where: T k is the temperature of partition k after power adjustment, and are the upper and lower limits of the zone temperature;

[0059] When the subproblem verification finds that the constraints are not satisfied, the cutting plane generation mechanism is triggered. The generated cutting plane is a linear inequality constraint, which is added to the main problem to limit the infeasible solution space: α T ΔP≤β

[0060] Where: α=[α1,α2,...,α K ] is the cutting plane coefficient vector, β is the cutting plane constant term;

[0061] S144, iterative update, add the cutting plane constraints generated by the subproblem to the main problem constraint set, and update the relaxed integer variable x in the main problem k Constraint range, update the number of iterations;

[0062] S145, iterative termination condition: when the difference between the objective function of the main problem and the sub-problem is less than 0.1% or the number of iterations is greater than 50, the iteration is terminated and the target adjustment value of each partition is output. The expression is: N iter >N max ,

[0063] Where: N iter is the current iteration number, N max is the maximum number of iterations, F1 is the function value of the main problem, F2 is the function value of the sub-problem, and ∈ is the difference between the objective functions of the main problem and the sub-problem.

[0064] Preferably, the step S2 specifically includes:

[0065] S21, construct a load adjustable domain model, including a second-order thermodynamic model of the temperature-controlled load and a dynamic model of the electric vehicle SOC;

[0066] The second-order thermodynamic model expression of the temperature control load is:

[0067] Where: is the indoor heat load, which is determined by the density of people and the heat generated by the equipment, K i is the heat transfer coefficient of the building envelope, C i is the building heat capacity, is the air conditioning power, η i is the air conditioning energy efficiency ratio, With T out are the indoor and outdoor temperatures, respectively;

[0068] The continuous time model is converted into a discrete time step using the Euler method, which is expressed as:

[0069]

[0070] The electric vehicle SOC dynamic model expression is:

[0071] Where: SOC i (t) is the battery state of charge at time t, is the charging power, is the battery capacity;

[0072] Constraints include:

[0073] When users are offline, they must meet the minimum SOC requirements:

[0074] Where: The lower limit of SOC when off-grid;

[0075] Temperature deviation constraint:

[0076] SOC safety margin constraint:

[0077] S22, with the goal of minimizing the adjustment cost, constructs a distributed optimization model with multiple constraints, and the objective function is:

[0078]

[0079] Where: is the adjustment cost coefficient, which is determined by the device type and user contract, and λ is the instruction tracking penalty coefficient, which ensures that the total adjustment amount is close to the instruction value;

[0080] Constraints include:

[0081] Adjustable power limit:

[0082] Where: The power adjustment ratio allowed by the user;

[0083] Thermal comfort constraints:

[0084] SOC recovery constraints:

[0085] S23, the parallel ADMM algorithm reduces the solution complexity through distributed computing, meeting the minute-level real-time requirements. An auxiliary variable z is introduced to represent the global consistency target. The iterative formula includes:

[0086] Local variable update:

[0087] Where: ρ is the penalty parameter, balancing the objective function and the consistency constraint, is the dual variable;

[0088] Global consistency update:

[0089] Dual variable update:

[0090] The iteration termination condition is:

[0091] Raw residuals: ∈ pri is the original residual threshold, ensuring that the local solution is close to the global consistency;

[0092] Dual residual: ||z k -z k-1 ||2<ò dual ,∈ dual is the dual residual threshold to ensure the stability of the global solution;

[0093] Maximum number of iterations: 100;

[0094] When the conditions are met, the global instructions are classified into load clusters.

[0095] Preferably, the step S3 specifically includes:

[0096] S31, calculating the upper and lower limits of real-time adjustable power according to the load adjustable power boundary model;

[0097] S32, using finite state machine logic to design four load states: ready, response, locked, and recovery, as well as state transition rules; filter available loads, and retain loads that are in the ready state and have power constraints that are not zero, based on the load states;

[0098] S33, dynamic priority sorting, dynamically adjusts the weight coefficient based on the system's real-time frequency deviation, and calculates the priority score based on the dynamic priority weight function. The higher the score, the higher the priority. A load response queue is generated, and high-scoring loads are called first. The queue length is dynamically adjusted;

[0099] S34, using FPGA hardware acceleration to achieve millisecond-level closed-loop control, including: switching command issuance, parallel execution and feedback verification, the feedback verification specifically includes: collecting real-time power change ΔP actual and the target power adjustment value P issued by the regional layer target , if |ΔP actual -ΔP target |>5%, then re-sort the queue.

[0100] The load adjustable power boundary model is specifically as follows:

[0101] The upper and lower limits of the real-time adjustable power of load i are defined as:

[0102] ΔP i max (t) = min(P i rated ,P i max (t)-P i current (t))

[0103] ΔP i min (t)=max(-P i rated ,P i min (t)-P i current (t))

[0104] Where: is the rated power, and are the upper and lower limits of power constraints, is the current actual power;

[0105] The dynamic priority weight function is specifically:

[0106] Define the priority score:

[0107] Where: To adjust the cost, the optimization result of the regional layer is provided. Δf(t) is the real-time frequency deviation of the system. α, β, γ are dynamic weight coefficients. α+β+γ=1. The larger the frequency deviation, the higher the γ weight. f threshold is the frequency deviation threshold.

[0108] Preferably, step S3 further includes a comfort level exceeding limit self-protection and elastic recovery mechanism, which monitors user comfort parameters. When the comfort parameters exceed the limit, the load is triggered to exit the response queue and enter the recovery mode, and the unavailable capacity is fed back to the regional layer, and the regional layer reallocates the adjustment task to other loads.

[0109] The comfort level exceeding limit judgment condition is: if the comfort level parameter of load i exceeds the allowable range, the self-protection is triggered. If T i (t)>T i max +δOR T i (t)<T i min -δ, then exit the response queue, where δ is a safety margin to prevent frequent switching;

[0110] The elastic recovery mechanism adopts a progressive power recovery strategy to avoid secondary power shocks caused by centralized recovery:

[0111] P i recover (t+1)=P i current (t)+k·(P i normal -P i current (t))

[0112] Where: k is the recovery rate coefficient, ensuring a smooth transition within 30 seconds, This is the power setting value under normal working conditions.

[0113] Compared with the prior art, the beneficial technical effects of the present invention are:

[0114] This application realizes hierarchical control of regional energy Internet by constructing a three-level hierarchical control architecture of "global-region-load" and combining it with the "time classification + space partition" collaborative mechanism.

[0115] The global layer is constructed through electrical coupling analysis and prediction technology, and a dynamic partitioning algorithm based on energy transmission margin is designed. The regional energy Internet is divided into multiple sub-regions with strong electrical coupling, and the equivalent adjustable capacity of flexible loads in each partition is aggregated. The global layer is optimized on an hourly time scale and generates partition-level power adjustment instructions, realizing cross-regional resource coordination and accurate reservation of spare capacity, solving the problems of high computational complexity and delayed adjustment instructions in traditional centralized optimization.

[0116] The regional layer constructs a thermal-economic hybrid constraint model based on the physical characteristics of flexible loads and the elasticity of user behavior. The parallel ADMM optimization algorithm is used to decompose the global instructions into load clusters within a minute time scale. By introducing user comfort recovery time constraints and flexible electricity price response functions, the dual goals of minimizing regulation costs and ensuring user satisfaction are achieved, breaking through the bottleneck of waste of regulation potential caused by load model simplification in traditional methods.

[0117] The load layer is constructed based on FPGA hardware acceleration and finite state machine logic, and a millisecond-level adjustable load priority switching mechanism is constructed. By calculating the load adjustable power boundary in real time and dynamically adjusting the response queue in combination with the system frequency deviation, precise regulation of flexible loads in seconds and self-protection against comfort level limits are achieved, which increases the traditional minute-level response speed by two orders of magnitude and significantly enhances the system stability in high-volatility scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0118] Figure 1 It is a schematic diagram of a three-level hierarchical control architecture;

[0119] Figure 2 Build a flow chart for the global layer;

[0120] Figure 3 Solve the flow chart for the parallel ADMM algorithm;

[0121] Figure 4 Constructing flow charts for the regional layer;

[0122] Figure 5 This is a flow chart for determining if comfort exceeds the limit;

[0123] Figure 6 This is the load layer control flow chart;

[0124] Figure 7 The figure compares the total latency of the embodiment with the traditional CPU solution. DETAILED DESCRIPTION

[0125] In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically limited. In the embodiments of the present application, all directional indications (such as up, down, left, right, front, back, top, bottom ...) are only used to explain the relative position relationship, movement, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication also changes accordingly. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or Internet of Things terminal that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or Internet of Things terminals.

[0126] In addition, references to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of such phrases in various places in the specification does not necessarily refer to the same embodiment, nor does it necessarily refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0127] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should be noted that the numerical values described in this application should not be fixed values. The numerical values are only example values and should not be limited to the scope of protection of this application. The relevant numerical data of this application can be determined according to the specific situation of the power grid.

[0128] The purpose of this invention is to propose a hierarchical and graded control method for a regional energy internet that takes flexible loads into account, in order to address the technical issues that existing hierarchical control technologies have difficulty adapting to the coordinated demands of high-proportion renewable energy volatility and second-level flexible load response, resulting in limited system regulation capabilities and low resource utilization. This application implements hierarchical and graded control of the regional energy internet by constructing a "global-regional-load" three-level hierarchical control architecture combined with a "time classification + spatial partitioning" collaborative mechanism.

[0129] like Figure 1 As shown, the present invention provides a hierarchical control method for a regional energy internet taking flexible loads into account, comprising the following steps:

[0130] S1: Construct a global layer. Using electrical coupling analysis and prediction technology, a dynamic partitioning algorithm based on energy transmission margin is designed to divide the regional energy internet into multiple sub-regions with strong electrical coupling. The equivalent adjustable capacity of the flexible loads in each sub-region is aggregated. The global layer is optimized on an hourly timescale, and an improved Benders decomposition algorithm is used to accelerate the solution and generate partition-level power regulation instructions.

[0131] S2: Build a regional layer and design a thermal-economic hybrid constraint model based on the physical characteristics of flexible loads and the elasticity of user behavior. This model introduces user comfort recovery time constraints and a flexible electricity price response function. Using a parallel ADMM optimization algorithm, the regional power regulation instructions issued by the global layer are decomposed into each flexible load cluster within a minute-level time scale.

[0132] S3 builds the load layer. Based on FPGA hardware acceleration and finite state machine logic, it constructs a millisecond-level adjustable load priority switching mechanism, calculates the load adjustable power boundary in real time, and dynamically adjusts the response queue based on the system frequency deviation.

[0133] Further, if Figure 2As shown, the global layer is constructed by dynamic partitioning and flexible capacity aggregation. Through dynamic partitioning and adjustable capacity aggregation, the complex energy Internet is decoupled into multiple autonomous control areas, and hourly regulation instructions are generated. Specifically, through electrical coupling analysis and digital twin prediction technology, a dynamic partitioning algorithm based on energy transmission margin is designed to divide the regional energy Internet into multiple electrically strongly coupled sub-areas, and aggregate the equivalent adjustable capacity of flexible loads in each partition. The global layer performs rolling optimization on an hourly time scale to generate partition-level power regulation instructions, realizing cross-regional resource coordination and precise reservation of spare capacity, and solving the problems of high computational complexity and lagging regulation instructions in traditional centralized optimization. It is specifically achieved through the following steps:

[0134] 1) Data preprocessing and coupling degree calculation

[0135] Input: Regional Energy Internet topology (node, line connection relationship), real-time line transmission power P ij (t), line capacity upper limit

[0136] Calculate the electrical coupling EC between all adjacent nodes ij (EC ij The closer it is to 1, the closer the transmission power of line ij is to the limit, and the stronger the coupling between nodes i and j.

[0137]

[0138] For each node i, generate its strongly coupled neighbor set:

[0139] Nb(i)={j|EC ij ≥γ}

[0140] Where γ is the threshold, which defaults to 0.6.

[0141] 2) Dynamic partition merging: strongly coupled nodes are merged into the same control area to reduce cross-region power interaction.

[0142] The processing flow is as follows:

[0143] (1) Perform initial partitioning, with each node forming a partition Ω k ={k},(k=1,2,...,N)

[0144] (2) Iterative merging rules:

[0145] Condition: If partition Ω a and Ω b There are common strong coupling nodes (i.e. And j∈Nb(i)), then merge into a new block Ω a+b .

[0146] Termination condition: There are no more partitions to merge.

[0147] (3) Partition validity check

[0148] Each partition must meet the lower limit of the total internal coupling degree:

[0149]

[0150] Where θ is the density threshold, which defaults to 0.5.

[0151] If not satisfied, the partition is split into smaller sub-regions.

[0152] (4) Output the final partition set {Ω1,Ω2,...,Ω K

[0153] 3) Partition-adjustable capacity aggregation

[0154] (1) Input the node set Ω of each partition k , flexible load parameter C i , temperature dead zone Response time User engagement α i (t)

[0155] (2) Calculation of adjustable power of single load:

[0156] For temperature-controlled loads, the adjustable power is the maximum power that can be adjusted per unit time within the allowable temperature deviation range. The calculation method is as follows:

[0157]

[0158] For electric vehicles:

[0159]

[0160] Where, Charging willingness coefficient for users, SOC i (t) is the state of charge of the electric vehicle, is the rated charge and discharge power of the electric vehicle.

[0161] (3) Dynamic correction of user engagement

[0162] In order to quantify users' willingness to respond in real time and avoid overestimation or underestimation of the adjustment potential caused by fixed coefficients, the correction model is designed as follows:

[0163]

[0164] Where: is the user's basic participation, which is determined by the user contract; ρ(t) is the real-time electricity price, ρbase is the benchmark electricity price; k p is the electricity price elasticity coefficient; is the temperature deviation so that the participation decreases when the actual temperature approaches the dead zone boundary.

[0165] (4) Partition capacity aggregation calculation

[0166] According to the partition node set Ω k With all loads With α i (t), polymerization is carried out based on the following formula:

[0167]

[0168] Where: β k The partition weight coefficient is dynamically adjusted according to the partition importance and defaults to 1.0.

[0169] The designed dynamic weight adjustment rule is: if the adjustment instruction completion rate of partition k in the past hour is less than 90%, then its weight is reduced:

[0170] β k =max(0.5,β k -0.1·(1-completion rate)

[0171] (5) Capacity verification

[0172] It is divided into physical constraint verification and economic verification. The physical constraint verification ensures that the aggregation result does not exceed the limit of the zoned power grid equipment:

[0173]

[0174] Where, Aggregate capacity, For critical line transmission capacity, is the transformer capacity

[0175] For economic verification, if the partition adjustment cost is higher than the threshold, it is marked as a "high-cost partition" and its adjustment amount is preferentially reduced in the global optimization.

[0176] 4) Global optimization instruction generation

[0177] The partition-level power adjustment instructions generated by the global layer can also be referred to as global optimization instructions or global instructions. The goal is to minimize the total cost of the system. The adjustment amount of each partition ΔP k As the decision variable, assign each partition adjustment instruction The objective function F1 is as follows:

[0178]

[0179] Among them, the cost of power generation That is, the quadratic function of thermal power cost; the penalty for energy abandonment Forecast output of renewable energy in zone k, is the actual renewable energy output, and γ is the penalty coefficient for wind and solar power curtailment (yuan / kWh). To smooth the item and avoid drastic fluctuations in the command, Adjustment instructions for the previous period.

[0180] Constraints include power balance constraints and adjustable capacity limits:

[0181]

[0182] Where, ΔP sys The total system deficit.

[0183] The Benders decomposition algorithm is used to accelerate the solution, and the relaxed integer variables and cutting plane generation mechanism are introduced to reduce the complexity of the main problem, accelerate convergence, and obtain the adjustment amount ΔP of each partition. k :

[0184] The main problem is set as a relaxed integer variable to solve the continuous partition power allocation.

[0185] The sub-problem is set to verify the feasibility of each partition constraint, generate a cutting plane (Cut) and return to the main problem.

[0186] Iteration termination condition: the difference between the objective function of the main problem and the sub-problem is <0.1% or the number of iterations is >50.

[0187] The specific solution process is as follows:

[0188] (1) Initialize parameters and set the initial partition power allocation value

[0189]

[0190] Where: is the initial power allocation value of partition k, ΔP total is the total power adjustment.

[0191] (2) Solve the main problem. Introduce slack integer variables to transform the original mixed integer programming problem into a continuous optimization problem. The objective function is F1. The constraints include power balance constraint and adjustable capacity limit. The introduced slack variable constraint ΔP k =x k , output the continuous partition power allocation value.

[0192] (3) Subproblem verification and cutting plane generation

[0193] The objective function of the subproblem is:

[0194]

[0195] Where, ΔP k is the power regulation of partition k, ΔP 1,k The power adjustment amount for partition k outputted by the main problem.

[0196] Verify the partition power adjustment ΔP k Whether the coupling constraint and user comfort constraint are satisfied. The coupling constraint is the mutual restriction relationship between the power regulation quantities of each partition, which is defined as follows:

[0197]

[0198] Where A ik is the sensitivity coefficient of partition k to constraint i, b i The upper limit of the capacity of constraint i.

[0199] The user comfort constraint is defined as:

[0200]

[0201] Where, T k is the temperature of partition k after power adjustment, and The upper and lower limits of the zone temperature.

[0202] When the subproblem verification finds that the constraints are not satisfied, the cutting plane generation mechanism is triggered. The generated cutting plane is a linear inequality constraint, which is added to the main problem to limit the infeasible solution space:

[0203] α T ΔP≤β

[0204] Where, α=[α1,α2,…,α K ] is the cutting plane coefficient vector, β is the cutting plane constant term.

[0205] (4) Iterative Update

[0206] Add the cutting plane constraints generated by the subproblem to the main problem constraint set and update the relaxed integer variable x in the main problem k The constraint range is updated and the number of iterations is updated.

[0207] (5) Termination condition judgment

[0208] When the difference between the objective function of the main problem and the sub-problem is less than 0.1% or the number of iterations is greater than 50, the iteration is terminated and the final power allocation value, i.e., the target adjustment value of each partition, is output.

[0209] N iter >N max

[0210]

[0211] Where N iter is the current iteration number, N max is the maximum number of iterations, F1 is the function value of the main problem, F2 is the function value of the sub-problem, ∈ is the difference between the objective function of the main problem and the sub-problem, and is taken as 0.1%.

[0212] Example

[0213] For example, a regional energy internet consists of three electrically tightly coupled sub-areas (Zone 1: Commercial, Zone 2: Industrial, and Zone 3: Residential). Global optimization instructions need to be generated on an hourly timescale, meeting the following operating conditions:

[0214] Total power regulation requirement: ΔPtotal = 120MW

[0215] The improved Benders decomposition algorithm is used to speed up the solution process as follows:

[0216] Solution to the first round main problem: ΔP = [45, 55, 20]MW

[0217] Verification failed: 0.85×45+0.92×55=93.65>90 → triggering cutting plane generation

[0218] Generate cutting plane: 0.85ΔP1+0.92ΔP2≤90

[0219] Solution to the main problem in round n: ΔP = [40, 50, 30]MW

[0220] Verification passed: 0.85×40+0.92×50=86≤90

[0221] The output is:

[0222] ΔP=[40,50,30]MW

[0223] 5) Instruction issuance and feedback correction

[0224] Output target adjustment amount for each partition (k=1,2,...,k)

[0225] The dynamic correction mechanism for settings is as follows:

[0226] The feedback signal is the actual adjustment amount of the regional layer

[0227] The revised rule is: if there are 3 consecutive cycles Repartition.

[0228] Further, if Figure 4As shown, the regional layer construction adopts distributed optimization and elastic constraint matching to convert the partition adjustment instructions issued by the global layer into Decompose it into each flexible load and achieve fast optimization at the minute level while meeting the user comfort and economic constraints. Specifically targeting the physical characteristics of flexible loads and the elasticity of user behavior, a thermal-economic hybrid constraint model is designed, and a parallel ADMM optimization algorithm is used to decompose global instructions into load clusters within a minute time scale. By introducing user comfort recovery time constraints and flexible electricity price response functions, the dual goals of minimizing regulation costs and ensuring user satisfaction are achieved, breaking through the bottleneck of wasted regulation potential caused by load model simplification in traditional methods. This is achieved through the following steps:

[0229] 1) Flexible load refined modeling

[0230] To establish a load-adjustable domain model that integrates physical characteristics and user behavior, and provide a constraint basis for optimization.

[0231] (1) Second-order thermodynamic model of temperature control load:

[0232] Considering the thermal inertia of the building, the temperature dynamic equation is established:

[0233]

[0234] Where: is the indoor heat load (kW), which is determined by the density of people and the heat generated by the equipment; K i Heat transfer coefficient of building envelope; C i is the building heat capacity; is the air conditioning power; η i is the air conditioning energy efficiency ratio, With T out Indoor and outdoor temperatures respectively.

[0235] The Euler method is used to convert the continuous time model into a discrete time step. The expression is:

[0236]

[0237] Electric vehicle SOC dynamic model:

[0238] Based on charging power and user travel requirements:

[0239]

[0240] Where SOC i (t) is the battery state of charge at time t; is the charging power, is the battery capacity.

[0241] Constraints:

[0242] When users are offline, they must meet the minimum SOC requirements:

[0243]

[0244] Where: It is the lower limit of SOC when off-grid.

[0245] The temperature deviation constraint must also be met:

[0246]

[0247] SOC safety margin constraint:

[0248]

[0249] 2) Optimize problem construction

[0250] Taking the power increase and decrease of load i as decision variables, a distributed optimization model with multiple constraints is constructed to minimize the adjustment cost and track the global instructions.

[0251] Objective function (minimize total cost)

[0252]

[0253] Where: is the adjustment cost coefficient (yuan / kWh), which is determined by the equipment type and user contract; λ is the instruction tracking penalty coefficient (default is 1000 yuan / kW 2 ), ensuring that the total adjustment amount is close to the command value.

[0254] Constraints:

[0255] Adjustable power limit:

[0256]

[0257] Where: The power adjustment ratio allowed by the user.

[0258] Thermal comfort constraints:

[0259]

[0260] SOC recovery constraints:

[0261]

[0262] 3) Parallel ADMM algorithm solution

[0263] like Figure 3The flowchart shown in the figure uses distributed computing to reduce solution complexity and meet minute-level real-time requirements. The global optimization problem is decomposed into N subproblems (each load is optimized independently), and an auxiliary variable z is introduced to represent the global consistency goal.

[0264] (1) Iterative formula

[0265] Local variable update (each load is solved in parallel):

[0266]

[0267] Where: ρ is the penalty parameter (default is 10), balancing the objective function and the consistency constraint. is the dual variable, and its initial value is 0.

[0268] Global consistency update:

[0269]

[0270] Dual variable update:

[0271]

[0272] (2) Termination conditions:

[0273] Set the algorithm termination conditions as follows. When the conditions are met, the global instructions are graded to the load cluster (minute time scale)

[0274] Raw residuals: ∈ pri The default value is 0.01, which ensures that the local solution is close to the global consistency.

[0275] Dual residual: ||z k -z k-1 ||2<ò dual ,∈ dual The default value is 0.001, which ensures the global solution is stable.

[0276] Maximum number of iterations: 100.

[0277] Further, if Figure 6The process shown in the figure constructs a load layer using priority response and fast closed-loop control. Based on FPGA hardware acceleration and finite state machine logic, a millisecond-level adjustable load priority switching mechanism is constructed. By calculating the load adjustable power boundary in real time and dynamically adjusting the response queue in combination with the system frequency deviation, precise regulation of flexible loads in seconds and self-protection against over-limit comfort are achieved, which increases the traditional minute-level response speed by two orders of magnitude and significantly enhances the system stability in high-volatility scenarios. Specifically, as the bottom layer of the hierarchical control architecture, the load layer needs to implement millisecond-level load switching and second-level frequency deviation control, including:

[0278] Dynamic priority sorting: Generate load response queues based on multiple objectives such as economy, regulation potential, and user comfort.

[0279] Real-time power boundary calculation: Dynamically calculates the adjustable power range by combining the physical characteristics of the load and user constraints.

[0280] FPGA hardware-accelerated closed-loop control: Through parallel computing and state machine logic, a millisecond-level closed-loop control system for instruction issuance, execution, and feedback is achieved.

[0281] Comfort level out-of-limit self-protection: monitor user comfort parameters (such as temperature and humidity) and trigger load exit or recovery strategies.

[0282] Further, the following steps are included:

[0283] Calculate the upper and lower limits of real-time adjustable power based on the load adjustable power boundary model;

[0284] Finite state machine logic is used to design the four states of the load: ready, response, locked, and recovery, as well as the state transition rules. Available loads are screened and, based on the load states, only those with a ready state and non-zero upper and lower power constraints are retained.

[0285] Dynamic priority sorting: dynamically adjusts the weight coefficient based on the system's real-time frequency deviation. The priority score is calculated based on the dynamic priority weight function. The higher the score, the higher the priority. A load response queue is generated, with high-scoring loads being called first. The queue length is dynamically adjusted.

[0286] Use FPGA hardware acceleration to achieve millisecond-level closed-loop control, including: switching command issuance, parallel execution and feedback verification. The feedback verification is specifically: collecting real-time power change ΔP actual and the target power adjustment value P issued by the regional layer target , if |ΔP actual -ΔP target |>5%, then re-sort the queue.

[0287] More specifically, the following formula is included:

[0288] (1) Adjustable power boundary model

[0289] Define the upper and lower limits of the real-time adjustable power of load i:

[0290] ΔP i max (t) = min(P i rated ,P i max (t)-P i current (t))

[0291] ΔP i min (t)=max(-P i rated ,P i min (t)-P i current (t))

[0292] Where: is the rated power, and are the upper and lower limits of power constraints, is the current actual power.

[0293] (2) Dynamic priority weight function

[0294] Define the priority score S i (t), taking into account the comprehensive economic performance, regulation potential and frequency deviation weights:

[0295]

[0296] Where: The cost is adjusted by the regional layer optimization result; Δf(t) is the real-time frequency deviation of the system; α, β, γ are dynamic weight coefficients, α+β+γ=1, the larger the frequency deviation, the higher the γ weight; f threshold is the frequency deviation threshold.

[0297] (3) Comfort level exceeding limit judgment conditions

[0298] like Figure 5 As shown, if the comfort parameter of load i is i (t) exceeds the allowable range and triggers self-protection:

[0299] If T i (t)>T i max +δOR T i (t)<T i min-δ, then exit the response queue. δ is a safety margin to prevent frequent switching.

[0300] Furthermore, the following processing flow is also included:

[0301] (1) Real-time data input and preprocessing

[0302] The input data includes the target power adjustment value P issued by the regional layer target ; System real-time frequency deviation Δf(t); load status information (including load power, temperature, etc.);

[0303] Perform preprocessing and use FPGA to parallel calculate the S of each load i (t), and And update it to the shared memory mapping table for subsequent module calls.

[0304] (2) Dynamic priority sorting and queue generation

[0305] Filter available loads and, based on the status of each load, retain the loads that are in the "Ready" state and and The load is not 0.

[0306] Calculate the priority score, dynamically adjust the weight coefficient according to the system real-time frequency deviation, and i (t) Calculate the score, the higher the score, the higher the priority.

[0307] Generate response queue, according to S i (t) The queues are sorted in descending order, with high-scoring loads being called first. The queue length is adjusted dynamically, ensuring that the total adjustable power is greater than the target value.

[0308] (3) FPGA hardware accelerated closed-loop control

[0309] This step mainly consists of three steps, as follows:

[0310] Instruction issuance: FPGA sends switching instructions to the load controller through the high-speed I / O interface. The instruction format is: {load ID, target power, execution timestamp}.

[0311] Parallel execution: Utilizes the FPGA's parallel pipeline architecture to process multiple load instructions simultaneously; single instruction processing latency ≤ 10ms, and overall response time ≤ 100ms.

[0312] Feedback verification: collect real-time power change ΔP actual , if |ΔP actual -ΔP target |>5%, then re-sort the queue.

[0313] (4) Finite state machine (FSM) logic design

[0314] Define the four states and conversion rules of load:

[0315] Ready: Can respond to commands and comfort parameters are normal.

[0316] Active: Power regulation is in progress and comfort levels are continuously monitored.

[0317] Locked: The comfort level exceeds the limit, forcibly exiting the queue and entering recovery mode.

[0318] Recovery: Slowly recover to a comfortable level with the lowest priority.

[0319] State transition logic:

[0320] Ready → Response: Receives command and

[0321] Response → Lock: T i (t) The limit is exceeded for 10 seconds (to prevent false triggering).

[0322] Lock → Restore: T i (t)Return to

[0323] Recovery → Ready: Recovery is complete and passes health check (power is stable for 30 seconds).

[0324] (5) Self-protection and elastic recovery mechanism

[0325] Set up an over-limit protection mechanism: when the load enters the "locked" state, the unavailable capacity is fed back to the regional layer; the regional layer reallocates the regulation tasks to other loads.

[0326] Adopt a gradual power recovery strategy to avoid secondary power shocks caused by centralized recovery:

[0327] P i recover (t+1)=P i current (t)+k·(P i normal -P i current (t))

[0328] Where k is the recovery rate coefficient, which ensures a smooth transition within 30 seconds and is set to 0.2; This is the power setting value under normal working conditions.

[0329] Example

[0330] The energy system of a commercial park needs to reduce the total load power by 500kW within 100ms when the grid frequency suddenly drops to 49.5Hz, below the nominal 50Hz. The system includes 200 air conditioners and 50 energy storage devices, each with a maximum adjustable power of 10kW. The millisecond-level response capability is verified through the following process:

[0331] The regional layer issues the target regulation amount: ΔPtarget = 500kW (reduction);

[0332] System frequency deviation: Δf = -0.5Hz;

[0333] Through FPGA pre-processing and parallel computing, it can complete the real-time adjustable power calculation of all loads within 1ms;

[0334] Dynamic priority sorting and queue generation are then performed to screen available loads. Loads exceeding temperature limits (T>26°C) or with a SOC <20% are eliminated, leaving 180 air conditioners and 45 energy storage units. The resource scores are calculated and a response queue is generated (queue update period: 10ms, implemented using FPGA hardware).

[0335] like Figure 7 The total latency shown in the figure compares it with the traditional CPU solution. The instruction issuance latency is approximately 2ms. During parallel execution, the energy storage responds by completing power adjustment within 10ms. The air conditioner responds by shutting down the compressor and reducing the power to the target value within 50ms, meaning the adjustment time is 50ms. The finite state machine state switching time, that is, the total latency from instruction issuance to state switching completion, is ≤2ms.

[0336] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in this application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are necessary for each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, and are not limitations. The above details do not limit this application to being implemented by adopting the above specific details. The above description disclosed is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be very obvious to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but to the widest scope consistent with the principles and novel features of this invention.

[0337] The above is only a preferred embodiment of the invention of this application and is not intended to limit the invention of this application. Any modifications, equivalent replacements, etc. made within the spirit and principles of the invention of this application should be included in the scope of protection of the invention of this application.

[0338] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A hierarchical control method for regional energy internet taking flexible loads into account, characterized in that: Including steps: S1: Construct a global layer. Using electrical coupling analysis and prediction technology, a dynamic partitioning algorithm based on energy transmission margin is designed to divide the regional energy internet into multiple sub-regions with strong electrical coupling. The equivalent adjustable capacity of the flexible loads in each sub-region is aggregated. The global layer is optimized on an hourly timescale, and an improved Benders decomposition algorithm is used to accelerate the solution and generate partition-level power regulation instructions. S2: Build a regional layer and design a thermal-economic hybrid constraint model based on the physical characteristics of flexible loads and the elasticity of user behavior. This model introduces user comfort recovery time constraints and a flexible electricity price response function. Using a parallel ADMM optimization algorithm, the regional power regulation instructions issued by the global layer are decomposed into each flexible load cluster within a minute-level time scale. S3 builds the load layer. Based on FPGA hardware acceleration and finite state machine logic, it constructs a millisecond-level adjustable load priority switching mechanism, calculates the load adjustable power boundary in real time, and dynamically adjusts the response queue based on the system frequency deviation.

2. A hierarchical control method for regional energy internet taking flexible loads into account according to claim 1, characterized in that: The step S1 specifically includes: S11, data preprocessing and electrical coupling calculation, calculate the electrical coupling EC between all adjacent nodes ij The expression is: Where: P ij (t) is the real-time line transmission power, is the upper limit of line capacity, EC ij The closer it is to 1, the closer the transmission power of line ij is to the limit, and the stronger the coupling between nodes i and j is; For each node i, generate a set of strongly coupled neighbors: Nb(i) = {j|EC ij ≥γ}where: γ is the threshold; S12, a dynamic partitioning algorithm is used to merge dynamic partitions, merging strongly coupled nodes into the same control area to reduce cross-area power interaction; S13, aggregate partition adjustable capacity; In step S14, an improved Benders decomposition algorithm is used to accelerate the solution and generate partition-level power adjustment instructions, wherein the objective function is set to minimize the total system cost: Where: Each partition adjustment amount ΔP k is the decision variable, Adjust instructions for each partition, is the cost of power generation, Punishment for abandonment. Forecast output of renewable energy in zone k, Contribute to actual renewable energy, is the regulation instruction of the previous period, γ is the penalty coefficient for wind and solar power abandonment, It is a smoothing term to avoid drastic fluctuations in instructions; The constraints include power balance constraint and adjustable capacity limit, which are expressed as: Where: ΔP sys is the total system shortfall; S15, issue instructions and correction feedback, and output the target adjustment amount for each partition Set up a dynamic correction mechanism: the feedback signal is the actual adjustment amount at the regional level The modified rule is that if there are 3 consecutive cycles Repartition.

3. A hierarchical control method for regional energy internet taking flexible loads into account according to claim 2, characterized in that: The step S12 is specifically as follows: S121, perform initial partitioning, each node forms a partition Ω k ={k},(k=1,2,...,N); S122, iterative merging rules: Condition: If partition Ω a and Ω b There are common strongly coupled nodes And j∈Nb(i), then merge into a new block Ω a+b ; Termination condition: There are no more partitions to merge; S123, partition validity check, each partition must meet the lower limit of the total internal coupling degree: Where: θ is the density threshold; If it is not satisfied, the partition is split into smaller sub-areas; S124, output the final partition set {Ω1,Ω2,...,Ω K }.

4. The hierarchical control method for regional energy internet taking flexible loads into account according to claim 2 is characterized in that: The step S13 is specifically as follows: S131, input load data, including: each partition node set Ω k , flexible load parameter C i , temperature dead zone Response time User engagement α i (t); S132, calculate the adjustable power of a single load, For temperature-controlled loads, the adjustable power is the maximum power that can be adjusted per unit time within the allowable temperature deviation range. The expression is: For electric vehicles, the expression is: Where: Charging willingness coefficient for users, SOC i (t) is the state of charge of the electric vehicle, is the rated charging and discharging power of the electric vehicle; S133, dynamically adjust user participation, the model expression is: Where: is the user's basic participation, which is determined by the user contract, ρ(t) is the real-time electricity price, ρ base is the base electricity price, k p is the electricity price elasticity coefficient, For temperature deviation, the participation is reduced when the actual temperature approaches the dead zone boundary; S134, partition capacity aggregation calculation, the expression is: Where: β k is the partition weight coefficient, which is dynamically adjusted according to the importance of the partition; The dynamic weight adjustment rule is: if the completion rate of the adjustment instruction of partition k in the past hour is less than 90%, then reduce its weight, the expression is: β k =max(0.5,β k -0.1·(1-completion rate)); S135: Perform capacity verification, including physical constraint verification and economic verification. The physical constraint verification ensures that the aggregation result does not exceed the limit of the zoned power grid equipment: Where: is the aggregate capacity, For critical line transmission capacity, is the transformer capacity; Economic Verification: If the partition adjustment cost is higher than the threshold, the high-cost partition is marked and its adjustment amount is preferentially reduced in the global optimization.

5. The hierarchical control method for regional energy internet taking flexible loads into account according to claim 2 is characterized in that: The improved Benders decomposition algorithm is used to accelerate the solution and generate the adjustment instructions as follows: S141, initializing parameters, setting the initial power allocation value of partition k to: Where: ΔP total is the total power adjustment; S142, the objective function of the main problem is F1, and the relaxed integer variable constraint ΔP is introduced. k =x k , output continuous partition power allocation value; S143, the sub-problem is set to verify the feasibility of each partition constraint, generate a cutting plane and return to the main problem. The objective function is: Where: ΔP k is the power regulation of partition k, ΔP 1,k The power adjustment amount of partition k outputted by the main problem; Verify the partition power adjustment ΔP k Whether the coupling constraint and user comfort constraint are satisfied. The coupling constraint is the mutual restriction relationship between the power regulation quantities of each partition, and the expression is: Where: A ik is the sensitivity coefficient of partition k to constraint i, b i is the upper capacity limit of constraint i; The user comfort constraint expression is: Where: T k is the temperature of partition k after power adjustment, and are the upper and lower limits of the zone temperature; When the subproblem verification finds that the constraints are not satisfied, the cutting plane generation mechanism is triggered. The generated cutting plane is a linear inequality constraint, which is added to the main problem to limit the infeasible solution space: α T ΔP≤β Where: α=[α1,α2,...,α K ] is the cutting plane coefficient vector, β is the cutting plane constant term; S144, iterative update, add the cutting plane constraints generated by the subproblem to the main problem constraint set, and update the relaxed integer variable x in the main problem k Constraint range, update the number of iterations; S145, iterative termination condition: When the difference between the objective function of the main problem and the sub-problem is less than 0.1% or the number of iterations is greater than 50, the iteration is terminated and the target adjustment value of each partition is output. The expression is: Where: N iter is the current iteration number, N max is the maximum number of iterations, F1 is the function value of the main problem, F2 is the function value of the sub-problem, and ∈ is the difference between the objective functions of the main problem and the sub-problem.

6. The hierarchical control method for regional energy internet taking flexible loads into account according to claim 1 is characterized in that: The step S2 specifically includes: S21, construct a load adjustable domain model, including a second-order thermodynamic model of the temperature-controlled load and a dynamic model of the electric vehicle SOC; The second-order thermodynamic model expression of the temperature control load is: Where: is the indoor heat load, which is determined by the density of people and the heat generated by the equipment, K i is the heat transfer coefficient of the building envelope, C i is the building heat capacity, is the air conditioning power, η i is the air conditioning energy efficiency ratio, With T out are the indoor and outdoor temperatures, respectively; The continuous time model is converted into a discrete time step using the Euler method, which is expressed as: The electric vehicle SOC dynamic model expression is: Where: SOC i (t) is the battery state of charge at time t, is the charging power, is the battery capacity; Constraints include: When users are offline, they must meet the minimum SOC requirements: Where: The lower limit of SOC when off-grid; Temperature deviation constraint: SOC safety margin constraint:

7. A hierarchical control method for regional energy internet taking flexible loads into account according to claim 6, characterized in that: The step S2 specifically further includes: S22, with the goal of minimizing the adjustment cost, constructs a distributed optimization model with multiple constraints, and the objective function is: Where: is the adjustment cost coefficient, which is determined by the device type and user contract, and λ is the instruction tracking penalty coefficient, which ensures that the total adjustment amount is close to the instruction value; Constraints include: Adjustable power limit: Where: The power adjustment ratio allowed by the user; Thermal comfort constraints: SOC recovery constraints: S23, the parallel ADMM algorithm reduces the solution complexity through distributed computing, meeting the minute-level real-time requirements. An auxiliary variable z is introduced to represent the global consistency target. The iterative formula includes: Local variable update: Where: ρ is the penalty parameter, balancing the objective function and the consistency constraint, is the dual variable; Global consistency update: Dual variable update: The iteration termination condition is: Raw residuals: ∈ pri is the original residual threshold, ensuring that the local solution is close to the global consistency; Dual residual: ∈ dual is the dual residual threshold to ensure the stability of the global solution; Maximum number of iterations: 100; When the conditions are met, the partition-level power adjustment instructions issued by the global layer are graded to the load cluster.

8. The hierarchical control method for regional energy internet taking flexible loads into account according to claim 1 is characterized in that: The step S3 specifically includes: S31, calculating the upper and lower limits of real-time adjustable power according to the load adjustable power boundary model; S32, using finite state machine logic to design four load states: ready, response, locked, and recovery, as well as state transition rules; filter available loads, and retain loads that are in the ready state and have power constraints that are not zero, based on the load states; S33, dynamic priority sorting, dynamically adjusts the weight coefficient based on the system's real-time frequency deviation, and calculates the priority score based on the dynamic priority weight function. The higher the score, the higher the priority. A load response queue is generated, and high-scoring loads are called first. The queue length is dynamically adjusted; S34, using FPGA hardware acceleration to achieve millisecond-level closed-loop control, including: switching command issuance, parallel execution and feedback verification, the feedback verification specifically includes: collecting real-time power change ΔP actual and the target power adjustment value P issued by the regional layer target , if |ΔP actual -ΔP target |>5%, then re-sort the queue.

9. The hierarchical control method for regional energy internet taking flexible loads into account according to claim 8, characterized in that: The load adjustable power boundary model is specifically: The upper and lower limits of the real-time adjustable power of load i are defined as: Where: is the rated power, and are the upper and lower limits of power constraints, is the current actual power; The dynamic priority weight function is specifically: Define the priority score: Where: To adjust the cost, the optimization result of the regional layer is provided. Δf(t) is the real-time frequency deviation of the system. α, β, γ are dynamic weight coefficients. α+β+γ=1. The larger the frequency deviation, the higher the γ weight. f threshold is the frequency deviation threshold.

10. The method for hierarchical and graded control of regional energy internet taking flexible load into account according to claim 8, characterized in that: Step S3 also includes a comfort level self-protection and elastic recovery mechanism, which monitors user comfort parameters. When the comfort parameters exceed the limit, the load is triggered to exit the response queue and enter recovery mode, and the unavailable capacity is fed back to the regional layer, and the regional layer reallocates the adjustment task to other loads. The comfort level exceeding limit judgment condition is: if the comfort level parameter of load i exceeds the allowable range, the self-protection is triggered. Then exit the response queue, where: δ is the safety margin to prevent frequent switching; The elastic recovery mechanism adopts a progressive power recovery strategy to avoid secondary power shocks caused by centralized recovery: Where: k is the recovery rate coefficient, ensuring a smooth transition within 30 seconds, This is the power setting value under normal working conditions.

Images