Multi-time-scale comprehensive demand response scheduling method for commercial building

Abstract

The invention provides a multi-time-scale comprehensive demand response scheduling method for a commercial building, and the method comprises the steps: firstly constructing a multi-time-scale comprehensive demand response scheduling frame of the commercial building, which comprises a day-ahead part and an intra-day part, and the intra-day part comprises a 15-minute level and a 1-minute level; secondly, aiming at responses of different time scales, constructing a main load model of the commercial building; and finally, on the basis of the two parts, constructing a multi-time scale comprehensive demand response scheduling model, and solving the model by adopting an improved grey wolf algorithm. The method aims at reasonably deploying load resources in commercial buildings, the day-ahead scheduling model can reduce daily operation cost, and the intra-day scheduling model can stabilize power fluctuation.

Description

[Technical Field]

[0001] This invention belongs to the field of integrated energy system technology, and specifically relates to a multi-timescale integrated demand response scheduling method for commercial buildings. [Background Technology]

[0002] In recent years, renewable energy power generation technology has developed rapidly in the commercial building sector. However, the randomness and intermittency of renewable energy power generation, as well as errors in load forecasting for commercial buildings, can easily cause frequent fluctuations in the power output of commercial buildings, posing challenges to their economic and safe operation. Without large-scale investment and renovation, optimizing the scheduling and energy management of resources within commercial buildings based on their power supply equipment configuration can reduce energy costs and effectively promote the safe use of distributed renewable energy sources.

[0003] Current research on optimal scheduling of commercial buildings suffers from problems such as strong reliance on forecasts and susceptibility to environmental factors. Day-ahead scheduling methods also fail to fully reflect the impact of uncertainties in renewable energy generation and load forecasting errors on the optimal operation of commercial buildings, easily leading to discrepancies between optimization results and actual operation. Because renewable energy output and load forecasting data contain errors, and these errors vary across different time scales, they affect the applicability and robustness of optimal scheduling results for commercial buildings. Therefore, this invention proposes a multi-time-scale integrated demand response scheduling method for commercial buildings, which overcomes the aforementioned deficiencies and shortcomings. A search revealed no prior art documents identical or similar to this invention. [Summary of the Invention]

[0004] The purpose of this invention is to propose a multi-timescale integrated demand response scheduling method for commercial buildings. This method overcomes the shortcomings of existing technologies and is a simple, easy-to-implement integrated energy demand response scheduling method. This method can realize the integrated demand response scheduling of various energy sources in commercial buildings at multiple timescales, providing support for the optimized scheduling of integrated energy systems in small-scale commercial buildings and improving the safety and stability of system operation.

[0005] The technical solution of this invention: A multi-timescale integrated demand response scheduling method for commercial buildings, such as... Figure 2 As shown, it includes the following steps:

[0006] Step 1: Construct an uncertain model of load and photovoltaic forecast results and a comprehensive demand response scheduling framework for commercial buildings across multiple time scales;

[0007] During each scheduling cycle, the scheduling platform obtains wind speed, light intensity, and outdoor temperature data from the weather station during the forecast period through data interaction; at the same time, it obtains indoor temperature, various loads, and rooftop photovoltaic operation data and information of commercial buildings through sensors installed in the commercial buildings, which serve as the basic data for realizing energy scheduling of commercial buildings.

[0008] The multi-timescale integrated demand response scheduling framework for commercial buildings in step 1 specifically includes two phases: day-ahead and intraday. Figure 3 As shown; the day-ahead phase is the economically optimal scheduling phase, used to reduce the operating costs of commercial buildings; the intraday phase is the adjustment phase, used to stabilize the power fluctuations of the tie lines of commercial buildings; due to the errors in the day-ahead forecasts of renewable energy and loads, the day-ahead phase refers to establishing a day-ahead uncertainty model based on multi-scenario technology; the intraday phase is based on establishing an intraday uncertainty model based on fuzzy numbers, thereby eliminating as many strong uncertainties as possible introduced by flexible loads and distributed power sources during the actual operation of commercial buildings.

[0009] The intraday phase includes two levels: 15-minute and 1-minute.

[0010] The specific aspects of establishing a day-ahead uncertainty model based on multi-scenario technology include:

[0011] Multi-scenario technology is a method for describing stochastic processes. It transforms load and photovoltaic forecast results containing uncertainties into a deterministic set of scenarios, enabling subsequent scheduling to take into account different error levels while simplifying calculations. Day-ahead uncertainty modeling includes two parts: scenario generation and scenario reduction.

[0012] (1.1) Scene generation:

[0013] First, parameter modeling is performed on load demand and photovoltaic output, and the actual value is expressed as the sum of the point prediction value and the random prediction error, as shown in formula (1):

[0014]

[0015] In the formula, and These represent the actual, predicted, and predicted values ​​of electrical, thermal, cooling, and gas loads and photovoltaic output, respectively.

[0016] Assuming that the day-ahead prediction error of multi-energy load follows a normal distribution with a mean of 0, and the day-ahead prediction error of photovoltaic output follows a TLS (Tlocation-scale) distribution, then the TLS distribution expression of the error ξ is shown in formula (2):

[0017]

[0018] In the formula: ν, σ, and μ are the shape parameter, scale parameter, and location parameter of the TLS distribution, respectively; Γ is the gamma function;

[0019] Based on the source-load uncertainty parameter model, deterministic scenarios are generated through sampling to simulate their randomness. Latin hypercube sampling (LHS) is used. The 24-hour power curve composed of sampled data represents the load or photovoltaic output scenario. Compared with Monte Carlo simulation, LHS can ensure that all sampling areas are covered by sampling points through stratified sampling. Therefore, the sampled values ​​of the random variables in LHS sampling are:

[0020]

[0021] In formula (3): Let X be a random variable m The nth sample value; F m U is the distribution function; n = (U+n-1) / N, which represents the random number of the nth subinterval after dividing the interval [0,1] into N equal parts, where U∈[0,1];

[0022] (1.2) Scene reduction:

[0023] Generally, the number of original scenes generated using LHS is large, and the calculation is complex, requiring reduction, i.e., using a small number of effective typical scenes to reflect the uncertainty characteristics. To break away from the dependence of traditional clustering algorithms on the number K of selected target scenes, this paper proposes an improved hierarchical K-means clustering algorithm based on the maximum distance method. The initial clustering value is selected by the maximum distance method, and the optimal cluster centers are obtained by the hierarchical K-means clustering algorithm. In the embodiment, photovoltaic power output is used as an example (others are similar):

[0024] (1.2.1) Suppose that a total of M days of valid source / load power data were acquired, then the original scene set for generating source / load power is P = [P1, P2, ..., P...]. M ], the power data vector for any scene is Pl=[p i,1 ,p i,2 ,...,p i,T Set the initial number of clusters K1.

[0025] (1.2.2) Select K1 initial cluster centers based on the maximum distance method;

[0026] The steps shown in (1.2.2) specifically include the following steps:

[0027] (1.2.2.1) Select the two scenes with the largest distance in the original scene set P of source / load power generated in step (1.2.1) as the initial cluster centers. The formula for calculating the distance d between the two scenes is shown in equation (4):

[0028]

[0029] In the formula, p i,t p j,t These represent the power values ​​of the i-th and j-th scenarios at time t, respectively.

[0030] (1.2.2.2) Among the remaining M-2 scenes in the original scene set P, the scene with the largest distance product with the two scenes that serve as the initial cluster centers is selected as the third cluster center. This process is repeated to obtain K1 initial cluster centers.

[0031] (1.2.3) Perform K-means clustering on the original scene set of source / load power, and group all scenes obtained in step (1.2.1) into the same cluster. To measure the nearest cluster center by the standard, let the iteration number l = 1, and calculate the value of the clustering measure function J in the l-th iteration. (l) The clustering measure function is calculated as follows:

[0032]

[0033] Where: M g Let g be the number of scenes in the g-th class; C is the h-th data vector in the g-th class; g Let g be the cluster center of the g-th class;

[0034] (1.2.4) After obtaining the measure in step (1.2.3), clustering is then performed. The specific steps are as follows:

[0035] (1.2.4.1) Utilization Calculate the radius between classes and select the class with the largest radius according to formula (6);

[0036]

[0037] (1.2.4.2) Select the two scenes with the largest distance from the largest radius determined in step (1.2.4.1) as the new cluster centers;

[0038] (1.2.4.3) Based on the new cluster centers determined in step (1.2.4.2), perform K-means clustering again, let l = l+1, and use formula (5) to calculate the (l+1)th clustering measure function value J. (l+1) .

[0039] (1.2.5) Define ε = (J (l) -J (l+1) ) / J (l) If ε is greater than the given threshold ε0, i.e. ε>ε0, then return to step (1.2.4) to continue iterating; otherwise, the algorithm ends and outputs the number of cluster centers and the clustering results.

[0040] The given threshold ε0 in step (1.2.5) can be set according to the change curve of the clustering measure function value.

[0041] (1.2.6) After obtaining the clustering results in step (1.2.5), assume there are G classes in the end, and the number of scenarios in each class is M. z,1 M z,2 ,...,M z,G The probability of each category is calculated using formula (7) based on the ratio of the number of photovoltaic scenarios in each category to the total number of photovoltaic scenarios in all categories.

[0042] ζ k =M z,k / (M z,1 +M z,2 +...+M z,G ), K = 1, 2, ..., G (7)

[0043] The specific steps for establishing an intraday uncertainty model based on fuzzy numbers include:

[0044] In intraday IES (Integrated Energy System) scheduling, weather forecast information is more accurate, and load energy consumption has certain time-series characteristics, so ultra-short-term photovoltaic output and load demand forecasts will be more accurate; however, considering that the robustness requirements are higher as the scheduling time approaches the operation point, uncertainty cannot be completely ignored; fuzzy mathematics theory describes and models fuzzy phenomena through precise mathematical means, and has high accuracy and simplicity, so an intraday uncertainty model for fuzzy chance-constrained programming is established.

[0045] The load demand or photovoltaic output in the form of point forecast is represented by triangular fuzzy variables (r1, r2, r3), and its membership function is shown in formula (8):

[0046]

[0047] In the formula, μ(x) is the membership function; r1, r2, r3 are membership parameters, satisfying r1 < r2 < r3, where r2 represents the most likely value of the variable; formulas (7) and (8) together constitute the intraday uncertainty model;

[0048] The triangular fuzzy number of intraday load or photovoltaic is shown in formula (9):

[0049]

[0050] In the formula, and The positive and negative prediction ratios are respectively, satisfying... It can be determined through historical prediction data; the three parts in parentheses represent the lower bound, median value, and upper bound of the triangular fuzzy parameter, respectively.

[0051] Step 2: Construct a primary load model for commercial buildings by assessing responses at different time scales;

[0052] Due to the thermal inertia of buildings, the heat exchange rate between indoors and outdoors is relatively slow. Meanwhile, electric vehicles (EVs) have a rapid charging and discharging response, making them suitable for responding to the energy management needs of commercial buildings within a short timescale. Therefore, this invention sets the air conditioner as a long-term scheduling device and the EV as a short-term scheduling device. The specific steps of step 2 include:

[0053] (2.1) Modeling of air conditioning load in commercial buildings:

[0054] Considering the relationship between air conditioning heat production and heat loss, the mathematical model of air conditioning load in commercial buildings is shown in formula (10):

[0055] Q4 = Q3 + Q6 + Q7 - (Q1 + Q2 + Q5) (10)

[0056] In the formula, Q1 is the convective heat transfer between the indoor wall and the air; Q2 is the heat loss due to infiltration through the window; Q3 is the radiative heat from outside light; Q4 is the increase in sensible heat of the building air per unit time; Q5 is the heat loss due to cold air intrusion / ventilation; Q6 is the heat exchange power between the indoor heat source and the indoor air; Q7 is the heat exchange power between the heating equipment and the indoor air.

[0057] The increase in sensible heat of the building air per unit time, Q4, can be obtained through formulas (11) and (12):

[0058] Q4 = C room dT in / dt=0.278c w ρ w V room dT in / dt (11)

[0059]

[0060] In the formula, C room A room refers to its heat capacity; R wall It is the thermal resistance of the interior walls; cw ρ is the specific heat capacity of air. w It is the density of air; V room It refers to the room's ventilation volume; P dh The heat dissipation power per unit area is taken as 3.8 W / m. 2 ;T in It is the indoor temperature, S room It refers to the room area; T HVAC It is the supply of cooling air for HVAC systems; T HVAC It refers to the supply air temperature of the HVAC system; T O Outdoor temperature; τ win For window penetration rate; A win It is a window area; I rad The light intensity through the window; N wall This is the total number of interior walls; T wall,i R represents the temperature of the i-th indoor wall surface. wall,i is the thermal resistance of the i-th wall in the room; L is the amount of outdoor air penetrating; V(t) is the ventilation volume during time period t, which can be approximately calculated using the air exchange rate method.

[0061] Then the power P of the commercial building load at time t CB (t) is shown in formula (13):

[0062] P CB (t)=P HVAC (t)+P other (t) (13)

[0063] In the formula, P HVAC (t) represents the power of the HVAC at time t; P other (t) represents the power of other loads at time t;

[0064] (2.2) Electric vehicle modeling:

[0065] The mathematical model of an electric vehicle includes equations (14)-(19); the probability density function of the EV battery capacity follows a gamma distribution, i.e.:

[0066]

[0067] 0.2≤SOC≤0.9 (16)

[0068] In the formula, E EV The remaining capacity of the electric vehicle battery is expressed in kWh; α is the shape parameter; β is the scale parameter. and These are the minimum and maximum capacity of the electric vehicle battery, respectively; SOC is the state of charge of the electric vehicle battery; to avoid overcharging and over-discharging of the battery, the reasonable range of SOC in the embodiment is [0.2, 0.9].

[0069] Electric vehicles have the characteristics of fast charging and discharging, and can be used as energy storage units in the energy management and operation optimization of commercial buildings. Therefore, at time t, the state of charge value of the s-th electric vehicle can be calculated according to formula (16):

[0070]

[0071] In the formula, and Let be the charging and discharging states of the s-th electric vehicle at time (t-1); and Let be the charging power and discharging power of the s-th electric vehicle at time (t-1), respectively. and Let be the charging and discharging efficiencies of the s-th electric vehicle at time t, and Δt be the time interval.

[0072] To ensure that the electric vehicle has sufficient battery capacity when leaving the office, i.e., enough for it to drive home, the following charging and discharging strategy is proposed: the State of Charge (SOC) of the electric vehicle leaving the office should be equal to the SOC of the electric vehicle arriving at the office; and at time t, the shortest duration t of the electric vehicle's last charging action. ch,end,s and the shortest duration t of the last discharge behavior of the electric vehicle dch,end,s They respectively satisfy the equality constraints of formulas (18) and (19);

[0073]

[0074] in, This represents the minimum state of charge at which an electric vehicle can drive home from the office.

[0075] Step 3: Based on the clustered scenarios obtained in Step 1 and the load and electric vehicle model in Step 2, construct a multi-time-scale integrated demand response scheduling model;

[0076] Step 3 specifically includes the following steps:

[0077] (3.1) Day-ahead demand response scheduling:

[0078] The main objective of demand response scheduling is to minimize daily operating costs while ensuring user temperature comfort. The objective function consists of two parts: economic cost and penalty for unsatisfactory temperature comfort. The economic cost includes the cost of purchasing electricity and the cost of using and maintaining equipment, as shown in formula (20).

[0079]

[0080] In the formula, n1 represents the number of scheduling cycles within a complete scheduling cycle of stage 1, and in this embodiment, n1 = 24; the first part of formula (20) is the cost of electricity purchased by the building system from the power distribution system; P ex (i) represents the power exchanged via the tie line between the building system and the power distribution system at time t, where power purchased from the power distribution system is defined as positive power, and power sold to the power distribution system is defined as negative power; C ph (t) and C se (t) represents the purchase price and sale price of electricity at time t, respectively; Part II of Equation (20) is the penalty function term affecting user temperature comfort, where γ is the user's sensitivity coefficient to temperature comfort. γ can be selected according to different user sensitivities. The larger γ is, the greater the loss caused by deviation from user temperature comfort, and vice versa; Part III is the usage and maintenance cost, P BESS (t), P PV (t), P WT (t), P HVAC (t), P HP (t) represents the charging and discharging power of the energy storage system at time t, and the output power of photovoltaic (PV), wind (W), heating, ventilation and air conditioning (HVAC), and heat pump (HP) at time t, respectively; C PV C WT C BESS C HVAC and C HP These represent the unit power consumption and maintenance cost of PV, W, battery energy storage system (BESS), HVAC, and HP, respectively.

[0081] The current demand response scheduling satisfies the constraints of formulas (21)-(25):

[0082] 1) Power balance constraints

[0083] P ex (t)+P PV (t)+P WT (t)+P BESS (t)=P CB (t) (21)

[0084] 2) Cold / Heat Balance Constraint

[0085] Q HVAC (t)+Q HP (t)=Q cl (t) (22)

[0086] Q HVAC (t)=P HVAC,h (t)×(1+COP HVAC ) (twenty three)

[0087] Q HP (t)=P HP (t)×(1+COP HP ) (twenty four)

[0088] In the formula, Q cl (t) represents the heating and cooling load at time t; Q HVAC (t) represents the cooling and heating output power of the HVAC system at time t; Q HP (t) represents the cooling and heating output power of HP at time t; COP HVAC For HVAC (Heating, Ventilation, Air Conditioning) energy efficiency ratio; COP HP This refers to HP's thermoelectric efficiency ratio.

[0089] 3) Indoor temperature comfort constraints

[0090]

[0091] In the formula, These represent the maximum and minimum indoor temperatures;

[0092] (3.2) Optimization and adjustment within 24 hours:

[0093] Based on the day-ahead economic optimal scheduling, the goal of hourly optimization adjustment is to effectively stabilize the deviation between the actual operating value and the day-ahead setpoint of the tie-line power, and to perform hourly optimal adjustment. The objective function is set as shown in formula (26):

[0094]

[0095] In the formula, The tie-line power setpoint for commercial buildings at time t′ under the day-ahead optimal scheduling scheme generated in Phase 1; P ex (t′) represents the actual power value of the connecting line of the commercial building at time t′; n2 represents the number of scheduling cycles in a complete scheduling cycle on the long time scale of stage 2. In this embodiment, n2 = 4. The constraints in this stage are still power balance constraints, cold and heat load balance constraints, building heat balance constraints, commercial building electricity purchase constraints, HVAC operation constraints, and indoor temperature constraints.

[0096] Step 4: Based on Step 3, the Modified Grey Wolf Optimizer (MGWO) algorithm is used to solve the above multi-timescale integrated demand response scheduling model.

[0097] The Grey Wolf Optimization Algorithm is a heuristic algorithm for solving optimization models. In this invention, individuals in the Grey Wolf Optimization Algorithm are considered as solutions to the integrated demand response scheduling model.

[0098] Based on the fitness of the gray wolf pack, i.e., the solution of the integrated demand response scheduling model, each individual gray wolf is assigned a different social status. The three optimal solutions are labeled α, β, and δ, representing the three leader gray wolves, who can guide other gray wolves to migrate to regions with better solutions. That is, other gray wolves adjust their positions according to the positions of the three leader gray wolves α, β, and δ. Under dimension p, the formula for updating the next position of gray wolf x under the guidance of wolves α, β, and δ is as follows:

[0099] X Ωx =X Ω -A1D Ω ,Ω=α,β,δ (27)

[0100] Among them, D Ω The calculation formula is D Ω =|C1X Ω -X Ω | represents the relative distance between gray wolf Ω and gray wolf x; C1 = 2r2 and A1 = 2ar1 - a are coefficients that adjust the exploration range and convergence speed, respectively, and r1 and r2 are random numbers between 0 and 1; a is the convergence factor.

[0101] In MGWO, the convergence factor is an important parameter, and its different update strategies can greatly affect the performance of the algorithm. Therefore, this paper proposes a new convergence factor update method based on exponential change, as shown in equation (28):

[0102] a = 2e -t / T (28)

[0103] Through iterative steps, the optimal fitness of the gray wolf pack can be obtained, which is the solution of the integrated demand response scheduling model. This solution can be used to achieve integrated demand response scheduling for commercial buildings across multiple time scales. [Attached Image Description]

[0104] Figure 1 This is a schematic diagram of the integrated energy system structure of a commercial building, which is part of the multi-timescale integrated demand response scheduling method for commercial buildings according to the present invention.

[0105] Figure 2 This is a schematic diagram of the overall process of a multi-timescale integrated demand response scheduling method for commercial buildings according to the present invention.

[0106] Figure 3This is a schematic diagram of a multi-timescale scheduling model for a multi-timescale integrated demand response scheduling method for commercial buildings, as described in this invention.

[0107] Figure 4 This is a flowchart illustrating the solution algorithm in the multi-timescale integrated demand response scheduling method for commercial buildings involved in this invention.

Detailed Implementation Methods

[0108] Embodiments of the present invention are described in detail below. Examples of these embodiments are given in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0109] A schematic diagram of the integrated energy system of the commercial building is shown below. Figure 1 As shown, a multi-timescale integrated demand response scheduling method for commercial buildings is proposed, such as... Figure 2 As shown, it includes the following steps:

[0110] Step 1: An uncertain model of load and photovoltaic forecast results and a comprehensive demand response scheduling framework for commercial buildings with multiple time scales were constructed, which includes two parts: day-ahead and intraday. The intraday part includes 15-minute and 1-minute levels.

[0111] Step 2: By assessing the response at different time scales, a primary load model for the commercial building was constructed;

[0112] Step 3: Based on the above two parts, a multi-timescale integrated demand response scheduling model was constructed.

[0113] Step 4: Based on Step 3, the Grey Wolf algorithm was improved, and the improved algorithm was used to solve the model to obtain the optimized scheduling result.

[0114] The specific steps of step 1 include:

[0115] During each scheduling cycle, the scheduling platform obtains data such as wind speed, solar irradiance, and outdoor temperature from weather stations via data exchange. Simultaneously, it acquires indoor temperature, various loads, and rooftop photovoltaic operation data and information from sensors installed in commercial buildings.

[0116] Energy dispatching methods for commercial buildings that consider demand response, such as Figure 3 As shown, this method consists of two phases. The first is the day-ahead economic optimal scheduling phase, which aims to reduce the operating costs of commercial buildings; the second is the intraday adjustment phase, which aims to stabilize the power fluctuations of the commercial building's tie lines.

[0117] Since there are certain errors in the day-ahead forecasts of renewable energy and load, this paper establishes a day-ahead uncertainty model based on multi-scenario technology in the day-ahead stage and an intraday uncertainty model based on fuzzy numbers in the intraday stage, so as to eliminate as much of the strong uncertainty introduced by flexible loads and distributed power sources as possible in the actual operation of commercial buildings.

[0118] Due to the thermal inertia of buildings, the heat exchange rate between indoors and outdoors is relatively slow. Meanwhile, electric vehicles (EVs) have a rapid charging and discharging response, making them suitable for responding to the energy management needs of commercial buildings within a short timescale. Therefore, this invention configures the air conditioner as a long-term scheduling device and the EV as a short-term scheduling device.

[0119] (1) Day-ahead uncertainty model based on multi-scenario technology

[0120] Multi-scenario technology is a method for describing stochastic processes. It transforms load and photovoltaic forecast results containing uncertainties into a deterministic set of scenarios, enabling subsequent scheduling to consider different error levels while simplifying computation. Day-ahead uncertainty modeling includes two parts: scenario generation and scenario reduction.

[0121] 1) Scene generation

[0122] First, parameter modeling is performed on load demand and photovoltaic output. Considering the commonly used point prediction method, this paper expresses the actual value as the sum of the point prediction value and the random prediction error.

[0123]

[0124] In the formula and These represent the actual, predicted, and predicted values ​​of electrical, thermal, cooling, and gas loads and photovoltaic output, respectively.

[0125] Based on existing research, we can assume that the day-ahead prediction errors of multi-energy loads all follow a normal distribution with a mean of 0, while the day-ahead prediction error of photovoltaic output follows a TLS (tlocation-scale) distribution. The TLS distribution expression for the error ξ is:

[0126]

[0127] In the formula: ν, σ, and μ are the shape parameter, scale parameter, and location parameter of the TLS distribution, respectively; Γ is the gamma function.

[0128] Based on the source-load uncertainty parameter model, deterministic scenarios are generated through sampling to simulate their randomness. Latin hypercube sampling (LHS) is used, and the sampled data forms a 24-hour power curve, which represents the load or photovoltaic output scenario. Compared with Monte Carlo simulation, LHS can ensure that all sampling areas are covered by sampling points through stratified sampling. Therefore, the sampled values ​​of the random variables in LHS sampling are:

[0129]

[0130] In the formula: Let X be a random variable m The nth sample value; F m U is the distribution function; n = (U+n-1) / N, which represents the random number of the nth subinterval after dividing the interval [0,1] into N equal parts, where U∈[0,1].

[0131] 2) Scene reduction.

[0132] Generally, the original scenes generated using LHS are numerous and computationally complex, necessitating reduction to use only a few effective, representative scenes to reflect the uncertainty characteristics. To overcome the dependence of traditional clustering algorithms on the number of selected target scenes K, this paper proposes an improved hierarchical K-means clustering algorithm based on the maximum distance method. The algorithm selects initial clustering values ​​using the maximum distance method and then obtains the optimal cluster centers using the hierarchical K-means clustering algorithm. For source / load power scenes, the scene reduction algorithm steps are as follows:

[0133] Step 1: Suppose that a total of M valid days of source / load power data have been acquired, then the original scene set for generating source / load power is P = [P1, P2, ..., P...]. M ], where the power data vector for any scene is P t =[p i,1 ,p i,2 ,...,p i,T Set the initial number of clusters K1.

[0134] Step 2: Select K1 initial cluster centers based on the maximum distance method. The specific steps are as follows.

[0135] Step 2.1: Select the two scenes with the largest distance in the generated original scene set P of source / load power as the initial cluster centers. The formula for calculating the scene distance d is:

[0136]

[0137] In the formula, p i,t p j,tThese represent the power values ​​of the i-th and j-th scenarios at time t, respectively.

[0138] Step 2.2: Among the remaining M-2 scenes, select the scene with the largest distance product between the first two initial scenes as the third cluster center; thus obtaining K1 initial cluster centers.

[0139] Step 3: Perform K-means clustering on the original scene set of source / load power, and group all scenes in (1.2.1) into the cluster with K-means clustering. To measure the nearest cluster center by the standard, let the iteration number l = 1, and calculate the value of the clustering measure function J in the l-th iteration. (l) The clustering measure function is calculated as follows:

[0140]

[0141] Where: M g Let g be the number of scenes in the g-th class; C is the h-th data vector in the g-th class; g Let g be the cluster center of the g-th class;

[0142] Step 4: After obtaining the measure in the above steps, clustering is performed. The specific steps are as follows:

[0143] Step 4.1: Utilize Calculate the radius between classes and select the class with the largest radius according to formula (6);

[0144]

[0145] Step 4.2: Select the two scenes with the largest distance in this class as the new cluster centers.

[0146] Step 4.3: Re-perform K-means clustering based on the cluster centers, let l = l + 1, and calculate the (l+1)th clustering measure function value J. (l+1) .

[0147] Step 5: Define ε = (J (l) -J (l+1) ) / J (l) If ε > ε0, then return to step 4 to continue iterating, where ε0 is a given threshold that can be set according to the change curve of the clustering measure function value; otherwise, the algorithm ends and outputs the number of cluster centers and the clustering results.

[0148] Step 6: After obtaining the clustering results, assume there are G clusters in total, and each cluster has M scenarios. z,1 M z,2 ,...,M z,G, calculate the probability ζ of each category according to the ratio of the number of photovoltaic scenarios in each category to the total number of photovoltaic scenarios in all categories k = M z,K / (M z,1 + M z,2 +... + M z,G ), K = 1, 2,..., G.

[0149] (2) Intra-day Uncertainty Model Based on Fuzzy Numbers

[0150] In the intra-day IES scheduling, information such as weather forecasts is more accurate, and at the same time, the load energy consumption has certain time-series characteristics. Therefore, the ultra-short-term photovoltaic output and load demand forecasts will be more accurate. However, considering that the robustness requirement is higher when approaching the operation time point in the scheduling, uncertainty cannot be completely ignored. Fuzzy mathematics theory describes and models fuzzy phenomena through precise mathematical means, with high accuracy and simplicity. Therefore, an intra-day uncertainty model for fuzzy chance-constrained programming is established.

[0151] The load demand and photovoltaic output in the form of point prediction can be represented by triangular fuzzy variables (r1, r2, r3), and its membership function is

[0152]

[0153] where: μ(x) is the membership function; r1, r2, r3 are membership degree parameters, satisfying r1 < r2 < r3, and r2 represents the most likely value of the variable. The specific steps of step 2 include:

[0154] Then the triangular fuzzy numbers of the intra-day load or photovoltaic are as shown in formula (8):

[0155]

[0156] where and are the positive and negative prediction proportionality coefficients respectively, satisfying which can be determined through historical prediction data; the three parts in the brackets represent the lower bound, middle value, and upper bound of the triangular fuzzy parameters respectively.

[0157] The specific steps of step 2 include:

[0158] (1) Modeling of Commercial Building Air Conditioning Load

[0159] Considering the relationship between air-conditioning heat production and heat loss, the mathematical model of commercial building air conditioning load is as shown in formula (9):

[0160] Q4 = Q3 + Q6 + Q7 - (Q1 + Q2 + Q5) (9)

[0161] In the formula, Q1 is the convective heat transfer between the indoor wall and the air; Q2 is the heat loss due to infiltration through the window; Q3 is the radiative heat from outside light; Q4 is the increase in sensible heat of the building air per unit time; Q5 is the heat loss due to cold air intrusion / ventilation; Q6 is the heat exchange power between the indoor heat source and the indoor air; and Q7 is the heat exchange power between the heating equipment and the indoor air.

[0162] Q4 = C room dT in / dt=0.278c w ρ w V room dT in / dt (10)

[0163]

[0164] In the formula, C room A room refers to its heat capacity; R wall It is the thermal resistance of the interior walls; c w ρ is the specific heat capacity of air. w It is the density of air; V room It refers to the room's ventilation volume; P dh The heat dissipation power per unit area is taken as 3.8 W / m. 2 ;T in It is the indoor temperature, S room It refers to the room area; T HVAC It is the supply of cooling air for HVAC systems; T HVAC It refers to the supply air temperature of the HVAC system; T O Outdoor temperature; τ win For window penetration rate; A win It is a window area; I rad The light intensity through the window; N wall This is the total number of interior walls; T wall,i R represents the temperature of the i-th indoor wall surface. wall,i is the thermal resistance of the i-th wall in the room; L is the amount of outdoor air penetrating; V(t) is the ventilation volume during time period t, which can be approximately calculated using the air exchange rate method.

[0165] P CB (t)=P HVAC (t)+P other (t) (12)

[0166] In the formula, P CB (t) Power of the commercial building load at time t; P HVAC (t) represents the power of the HVAC at time t; P other (t) represents the power of other loads at time t.

[0167] (2) Electric vehicle modeling

[0168] The mathematical model of an electric vehicle includes equations (13)-(18). The probability density function of the EV battery capacity follows a gamma distribution:

[0169]

[0170] 0.2≤SOC≤0.9 (15)

[0171] In the formula, E EV The remaining capacity of the electric vehicle battery is expressed in kWh; α is the shape parameter; β is the scale parameter. and These are the minimum and maximum capacity of the electric vehicle battery, respectively; SOC is the state of charge of the electric vehicle battery; to avoid overcharging and over-discharging of the battery, the reasonable range of SOC in this section is [0.2, 0.9].

[0172] Electric vehicles, characterized by rapid charging and discharging, can serve as energy storage units in the energy management and operational optimization of commercial buildings. At time t, the state of charge (SOC) of the i-th electric vehicle can be calculated using the following formula.

[0173]

[0174] In the formula, and Let be the charging and discharging states of the s-th electric vehicle at time (t-1); and Let be the charging power and discharging power of the s-th electric vehicle at time (t-1), respectively. and Let be the charging and discharging efficiencies of the s-th electric vehicle at time t, and Δt be the time interval.

[0175] To ensure that the electric vehicle has sufficient battery capacity when leaving the office, i.e., when it can drive home, this section proposes the following charging and discharging strategy: the State of Charge (SOC) of the electric vehicle leaving the office should be equal to the SOC of the electric vehicle arriving at the office. The shortest duration t of the electric vehicle's last charging action at time t is also considered. ch,end,s and the shortest duration t of the last discharge behavior of the electric vehicle dch,end,s They respectively satisfy the equality constraints of formulas (17) and (18).

[0176]

[0177] in, This represents the minimum state of charge at which an electric vehicle can drive home from the office.

[0178] The specific steps of step 3 include:

[0179] (1) Day-ahead demand response scheduling

[0180] The primary objective is to minimize daily operating costs while ensuring user temperature comfort. The objective function consists of two parts: economic costs and penalties for unsatisfactory temperature comfort. Economic costs include the cost of purchasing electricity and the costs of using and maintaining the equipment.

[0181]

[0182] In the formula, n1 represents the number of scheduling cycles within a complete scheduling cycle of Phase 1, and in this paper, n1 = 24; the first part of the above formula is the cost of electricity purchased by the building system from the distribution system; P ex (i) represents the power exchanged via the tie line between the building system and the power distribution system at time t. Power purchased from the power distribution system is positive power, and power sold to the power distribution system is negative power. C ph (t) and C se (t) represents the purchase price and sale price of electricity at time t, respectively. Part II of the above equation represents the penalty function term affecting user temperature comfort, where γ is the user's sensitivity coefficient to temperature comfort. γ can be selected based on different user sensitivities; the larger γ is, the greater the loss caused by deviation from user temperature comfort, and vice versa. Part III represents the usage and maintenance costs, P... BESS (t), P PV (t), P WT (t), P HVAC (t), P HP (t) represents the charging and discharging power of the energy storage system at time t, and the output power of photovoltaic (PV), wind (W), heating, ventilation and air conditioning (HVAC), and heat pump (HP) at time t, respectively; C PV C WT C BESS C HVAC and C HP These represent the unit power consumption and maintenance cost of PV, W, battery energy storage system (BESS), HVAC, and HP, respectively.

[0183] 1) Power balance constraints

[0184] P ex (t)+P PV (t)+P WT (t)+P BESS (t)=PCB (t)(20)

[0185] 2) Cold / Heat Balance Constraint

[0186] Q HVAC (t)+Q HP (t)=Q cl (t) (21)

[0187] Q HVAC (t)=P HVAC,h (t)×(1+COP HVAC ) (twenty two)

[0188] Q HP (t)=P HP (t)×(1+COP HP ) (twenty three)

[0189] In the formula, Q cl (t) represents the heating and cooling load at time t; Q HVAC (t) represents the cooling and heating output power of the HVAC system at time t; Q HP (t) represents the cooling and heating output power of HP at time t; COP HVAC For HVAC (Heating, Ventilation, Air Conditioning) energy efficiency ratio; COP HP This refers to HP's thermoelectric efficiency ratio.

[0190] 3) Indoor temperature comfort constraints

[0191]

[0192] In the formula, These represent the maximum and minimum indoor temperatures.

[0193] (2) Optimization and adjustment within hours

[0194] Based on the day-ahead economic optimal scheduling, the goal of hourly optimization is to effectively stabilize the deviation between the actual operating value and the day-ahead setpoint of the tie-line power, and to perform hourly optimal adjustments. The objective function is as follows:

[0195]

[0196] In the formula, P represents the tie-line power setpoint for commercial buildings at time t′ under the day-ahead optimal scheduling scheme generated in Phase 1. ex (t′) represents the actual power value of the connecting line of the commercial building at time t′. n2 is the number of scheduling cycles in a complete scheduling cycle on the long time scale of stage 2. In this embodiment, n2 = 4. The constraints in this stage are still power balance constraints, heating and cooling load balance constraints, building heat balance constraints, commercial building electricity purchase constraints, HVAC operation constraints, and indoor temperature constraints, etc.

[0197] Step 4: Based on Step 3, the improved Grey Wolf algorithm is used to solve the above multi-timescale integrated demand response scheduling model, such as... Figure 4 As shown. The specific steps of step 4 include:

[0198] The Grey Wolf Optimization Algorithm is a heuristic algorithm for solving optimization models. In this invention, individuals in the Grey Wolf Optimization Algorithm are considered as solutions to the integrated demand response scheduling model.

[0199] Based on the fitness of the wolf pack (i.e., the solution of the integrated demand response scheduling model), different social statuses are assigned, and the three optimal solutions are labeled as a, β, and δ, respectively. These leader wolves guide other wolves to migrate to regions with better solutions. The following details how each leader role affects position updates.

[0200] Other gray wolves tend to adjust their positions based on the positions of α, β, and δ. In dimension p, the formula for updating the next position of gray wolf x under the guidance of α, β, and δ is as follows:

[0201] X Ωx =X Ω -A1D Ω ,Ω=a,β,δ (27)

[0202] Among them, D Ω The calculation formula is D Ω =|C1X Ω -X Ω | represents the relative distance between Ω and the gray wolf x; C1=2r2 and A1=2ar1-a are coefficients that adjust the exploration range and convergence speed, respectively, and r1 and r2 are random numbers between 0 and 1. a is the convergence factor.

[0203] In MGWO, different update strategies for important parameters can greatly affect the performance of the algorithm. Therefore, this paper proposes a new convergence factor update method based on exponential change, as shown in the equation.

[0204] a = 2e -t / T (27)

[0205] Through gradual iteration, the optimal fitness of the gray wolf pack is obtained as the solution for the comprehensive demand response scheduling model.

Claims

1. A multi-timescale integrated demand response scheduling method for commercial buildings, characterized in that... It includes the following steps: Step 1: Construct an uncertain model of load and photovoltaic forecast results and a comprehensive demand response scheduling framework for commercial buildings across multiple time scales; Step 2: Construct a primary load model for commercial buildings by assessing responses at different time scales; Step 3: Based on the clustered scenarios obtained in Step 1 and the load and electric vehicle model in Step 2, construct a multi-time-scale integrated demand response scheduling model; Step 4: Based on Step 3, use the improved Grey Wolf algorithm to solve the multi-timescale integrated demand response scheduling model.

2. The integrated demand response scheduling method for commercial buildings across multiple time scales as described in claim 1, characterized in that... The multi-timescale integrated demand response scheduling framework for commercial buildings in step 1 specifically includes two stages: the day-ahead stage and the intraday stage. The day-ahead stage is the economically optimal scheduling stage, which is used to reduce the operating costs of commercial buildings. It refers to establishing a day-ahead uncertainty model based on multi-scenario technology. The intraday stage is the adjustment stage, which is used to stabilize the power fluctuations of the connecting lines of commercial buildings. It is based on establishing an intraday uncertainty model based on fuzzy numbers.

3. The integrated demand response scheduling method for commercial buildings across multiple time scales as described in claim 2, characterized in that... The intraday phase includes two levels: 15-minute and 1-minute.

4. The integrated demand response scheduling method for commercial buildings across multiple time scales according to claim 2, characterized in that... The specific content of establishing a day-ahead uncertainty model based on multi-scenario technology includes: (1.1) Scene generation: First, parameter modeling is performed on load demand and photovoltaic output, and the actual value is expressed as the sum of the point prediction value and the random prediction error, as shown in formula (1): In the formula, and These represent the actual, predicted, and predicted values ​​of electrical, thermal, cooling, and gas loads and photovoltaic output, respectively. Assuming that the day-ahead prediction error of multi-energy load follows a normal distribution with a mean of 0, and the day-ahead prediction error of photovoltaic output follows a TLS (Tlocation-scale) distribution, then the TLS distribution expression of the error ξ is shown in formula (2): In the formula: ν, σ, and μ are the shape parameter, scale parameter, and location parameter of the TLS distribution, respectively; Γ is the gamma function; Based on the source-load uncertain parameter model, the randomness of the deterministic scenario is simulated by sampling. The sampled values ​​of the random variables using Latin hypercube sampling (LHS) are: In formula (3): Let X be a random variable m The nth sample value; F m U is the distribution function; n = (U+n-1) / N, which represents the random number of the nth subinterval after dividing the interval [0,1] into N equal parts, where U∈[0,1]; (1.2) Scene reduction: (1.2.1) Suppose that a total of M days of valid source / load power data were acquired, then the original scene set for generating source / load power is P = [P1, P2, ..., P...]. M ], the power data vector for any scene is Pl=[p i,1 ,p i,2 ,...,p i,T Set the initial number of clusters K1; (1.2.2) Select K1 initial cluster centers based on the maximum distance method; (1.2.3) Perform K-means clustering on the original scene set of source / load power, and group all scenes obtained in step (1.2.1) into the same cluster. To measure the nearest cluster center by the standard, let the iteration number l = 1, and calculate the value of the clustering measure function J in the l-th iteration. (l) The clustering measure function is calculated as follows: Where: M g Let g be the number of scenes in the g-th class; C is the h-th data vector in the g-th class; g Let g be the cluster center of the g-th class; (1.2.4) After obtaining the measure in step (1.2.3), clustering is then performed. The specific steps are as follows: (1.2.4.1) Utilization Calculate the radius between classes and select the class with the largest radius according to formula (6); (1.2.4.2) Select the two scenes with the largest distance from the largest radius determined in step (1.2.4.1) as the new cluster centers; (1.2.4.3) Based on the new cluster centers determined in step (1.2.4.2), perform K-means clustering again, let l = l+1, and use formula (5) to calculate the (l+1)th clustering measure function value J. (l+1) ; (1.2.5) Define ε = (J (l) -J (l+1) ) / J (l) If ε is greater than the given threshold ε0, i.e. ε>ε0, then return to step (1.2.4) to continue iterating; otherwise, the algorithm ends and outputs the number of cluster centers and the clustering results. (1.2.6) After obtaining the clustering results in step (1.2.5), assume there are G classes in the end, and the number of scenarios in each class is M. z,1 M z,2 ,...,M z,G The probability of each category is calculated using formula (7) based on the ratio of the number of photovoltaic scenarios in each category to the total number of photovoltaic scenarios in all categories. ζ k =M z,k / (M z,1 +M z,2 +...+M z,G ),K=1,2,...,G (7)。 5. The integrated demand response scheduling method for commercial buildings across multiple time scales according to claim 4, characterized in that... Step (1.2.2) specifically includes the following steps: (1.2.2.1) Select the two scenes with the largest distance in the original scene set P of source / load power generated in step (1.2.1) as the initial cluster centers. The formula for calculating the distance d between the two scenes is shown in equation (4): In the formula, p i,t p j,t These represent the power values ​​of the i-th and j-th scenarios at time t, respectively. (1.2.2.2) Among the remaining M-2 scenes in the original scene set P, the scene with the largest distance product with the two scenes that serve as the initial cluster centers is selected as the third cluster center. This process is repeated to obtain K1 initial cluster centers.

6. The integrated demand response scheduling method for commercial buildings across multiple time scales according to claim 4, characterized in that... The given threshold ε0 in step (1.2.5) can be set according to the change curve of the clustering measure function value.

7. The integrated demand response scheduling method for commercial buildings across multiple time scales according to claim 2, characterized in that... The specific steps for establishing an intraday uncertainty model based on fuzzy numbers include: Use triangular fuzzy variables (r1, r2, r3) to represent the load demand or photovoltaic output in the form of point prediction, and its membership function is shown in Equation (8): In the formula, μ(x) is the membership function; r1, r2, r3 are membership degree parameters, satisfying r1 < r2 < r3, and r2 represents the most likely value of the variable; Equations (7) and (8) together constitute the intraday uncertainty model; Then the triangular fuzzy number of intraday load or photovoltaic is shown in Equation (9): In the formula, and The positive and negative prediction ratios are respectively, satisfying... It can be determined through historical prediction data; the three parts in parentheses represent the lower bound, median value, and upper bound of the triangular fuzzy parameter, respectively.

8. The integrated demand response scheduling method for commercial buildings across multiple time scales according to claim 1, characterized in that... The specific steps of Step 2 include: (2.1) Modeling of commercial building air-conditioning load: Considering the relationship between air-conditioning heat production and heat loss, the mathematical model of commercial building air-conditioning load is shown in Equation (10): Q4 = Q3 + Q6 + Q7 - (Q1 + Q2 + Q5) (10) In the formula, Q1 is the convective heat transfer between the indoor wall surface and air; Q2 is the infiltration heat loss of the window; Q3 is the radiant heat of the outdoor light; Q4 is the sensible heat increment of the building air per unit time; Q5 is the cold air intrusion / ventilation heat loss; Q6 is the heat transfer power between the indoor heat source and indoor air; Q7 is the heat transfer power between the heating equipment and indoor air; Among them, the sensible heat increment Q4 of the building air per unit time can be obtained through Equations (11) and (12): Q4=C room dT in / dt=0.278c w ρ w V room dT in / dt (11) In the formula, C room A room refers to its heat capacity; R wall It is the thermal resistance of the interior walls; c w ρ is the specific heat capacity of air. w It is the density of air; V room It refers to the room's ventilation volume; P dh The heat dissipation power per unit area is taken as 3.8 W / m. 2 ;T in It is the indoor temperature, S room It refers to the room area; T HVAC It is the supply of cooling air for HVAC systems; T HVAC It refers to the supply air temperature of the HVAC system; T O Outdoor temperature; T win For window penetration rate; A win It is a window area; I rad The light intensity through the window; N wall This is the total number of interior walls; T wall,i R represents the temperature of the i-th indoor wall surface. wall,i is the thermal resistance of the i-th wall in the room; L is the amount of outdoor air penetrating; V(t) is the ventilation volume during time period t, which can be approximately calculated using the air exchange rate method. Then the power P of the commercial building load at time t CB (t) is shown in formula (13): P CB (t)=P HVAC (t)+P other (t) (13) In the formula, P HVAC (t) represents the power of the HVAC at time t; P other (t) represents the power of other loads at time t; (2.2) Modeling of electric vehicles: The mathematical model of electric vehicles includes Equations (14)-(19); the probability density function of the EV battery capacity follows the gamma distribution, that is: 0.2 ≤ SOC ≤ 0.9 (16) In the formula, E EV The remaining capacity of the electric vehicle battery is expressed in kWh; α is the shape parameter; β is the scale parameter. and These are the minimum and maximum capacity of the electric vehicle battery, respectively; SOC is the state of charge of the electric vehicle battery; to avoid overcharging and over-discharging of the battery, the reasonable range of SOC in the embodiment is [0.2, 0.9]. Electric vehicles have the characteristics of fast charging and discharging and can participate in the energy management and operation optimization of commercial buildings as energy storage units. Therefore, at time t, the state of charge value of the s-th electric vehicle can be calculated according to Equation (16): In the formula, and Let be the charging and discharging states of the s-th electric vehicle at time (t-1); and Let be the charging power and discharging power of the s-th electric vehicle at time (t-1), respectively. and Let be the charging and discharging efficiencies of the s-th electric vehicle at time t, and Δt be the time interval. To ensure that the electric vehicle has sufficient battery capacity when leaving the office, i.e., when it can drive home, the charging and discharging strategy is as follows: the SOC of the electric vehicle leaving the office should be equal to the SOC of the electric vehicle arriving at the office; and at time t, the shortest duration t of the electric vehicle's last charging action. ch,end,s and the shortest duration t of the last discharge behavior of the electric vehicle dch,end,s They respectively satisfy the equality constraints of formulas (18) and (19); in, This represents the minimum state of charge at which an electric vehicle can drive home from the office.

9. The integrated demand response scheduling method for commercial buildings across multiple time scales according to claim 1, characterized in that... The specific steps of Step 3 include the following steps: (3.1) Day-ahead demand response scheduling: The main goal of day-ahead demand response scheduling is to minimize the daily operating cost on the basis of ensuring user temperature comfort. The objective function consists of two parts: the economic cost and the penalty caused by the dissatisfaction of temperature comfort. Among them, the economic cost includes the cost of purchasing electricity and the use and maintenance costs of equipment, as shown in Equation (20): In the formula, n1 represents the number of scheduling cycles within a complete scheduling cycle of stage 1, and in this embodiment, n1 = 24; the first part of formula (20) is the cost of electricity purchased by the building system from the power distribution system; P ex (i) represents the power exchanged via the tie line between the building system and the power distribution system at time t, where power purchased from the power distribution system is defined as positive power, and power sold to the power distribution system is defined as negative power; C ph (t) and C se (t) represents the purchase price and sale price of electricity at time t, respectively; Part II of Equation (20) is the penalty function term affecting user temperature comfort, where γ is the user's sensitivity coefficient to temperature comfort. γ can be selected according to different user sensitivities. The larger γ is, the greater the loss caused by deviation from user temperature comfort, and vice versa; Part III is the usage and maintenance cost, P BESS (t), P PV (t), P WT (t), P HVAC (t), P HP (t) represents the charging and discharging power of the energy storage system at time t, and the output power of photovoltaic (PV), wind (W), heating, ventilation and air conditioning (HVAC), and heat pump (HP) at time t, respectively; C PV C WT C BESS C HVAC and C HP These represent the unit power consumption and maintenance cost of PV, W, battery energy storage system (BESS), HVAC, and HP, respectively. The day-ahead demand response scheduling satisfies the constraint conditions as shown in Equations (21)-(25): 1) Power balance constraint P ex (t)+P PV (t)+P WT (t)+P BESS (t)=P CB (t) (21) 2) Cold / hot balance constraint Q HVAC (t)+Q HP (t)=Q cl (t) (22) Q HVAC (t)=P HVAC,h (t)×(1+COP HVAC ) (23) Q HP (t)=P HP (t)×(1+COP HP ) (24) In the formula, Q cl (t) represents the heating and cooling load at time t; Q HVAC (t) represents the cooling and heating output power of the HVAC system at time t; Q HP (t) represents the cooling and heating output power of HP at time t; COP HVAC For HVAC (Heating, Ventilation, Air Conditioning) energy efficiency ratio; COP HP This refers to HP's thermoelectric efficiency ratio. 3) Indoor temperature comfort constraint In the formula, These represent the maximum and minimum indoor temperatures; (3.2) Hourly optimization adjustment: On the basis of the day-ahead economic optimal scheduling, the goal of hourly optimization adjustment is to effectively stabilize the deviation between the actual operation value and the day-ahead set value of tie-line power, and perform hourly optimal adjustment. The objective function is set as shown in Equation (26): In the formula, The tie-line power setpoint for commercial buildings at time t′ under the day-ahead optimal scheduling scheme generated in Phase 1; P ex (t′) represents the actual power value of the connecting line of the commercial building at time t′; n2 represents the number of scheduling cycles in a complete scheduling cycle on the long time scale of stage 2. In this embodiment, n2 = 4. The constraints in this stage are still power balance constraints, cold and heat load balance constraints, building heat balance constraints, commercial building electricity purchase constraints, HVAC operation constraints, and indoor temperature constraints.

10. The integrated demand response scheduling method for commercial buildings across multiple time scales according to claim 1, characterized in that... The specific steps of Step 4 include: According to the fitness of the grey wolf population, that is, the solution of the comprehensive demand response scheduling model, each grey wolf individual is given a different social status. The three optimal solutions are respectively marked as α, β, δ, representing three leading grey wolves, which can guide other grey wolves to migrate to the more optimal solution area, that is, other grey wolves adjust their positions according to the positions of the three leading grey wolves α, β, δ. Under dimension p, the update formula of the next position of grey wolf x under the guidance of wolf a, wolf β, and wolf δ is as follows: X Ωx =X Ω -A1D Ω ,Ω=a,b,d (27) Among them, D Ω The calculation formula is D Ω =|C1X Ω -X Ω | represents the relative distance between gray wolf Ω and gray wolf x; C1 = 2r2 and A1 = 2ar1 - a are coefficients that adjust the exploration range and convergence speed, respectively, and r1 and r2 are random numbers between 0 and 1; a is the convergence factor. In MGWO, the convergence factor is an important parameter, and its different update strategies can greatly affect the performance of the algorithm. Therefore, this paper proposes a new convergence factor update method based on exponential change, as shown in equation (28): a=2e -t / T (28) Through iterative steps, the optimal fitness of the gray wolf pack can be obtained, which is the solution of the integrated demand response scheduling model. This solution can be used to achieve integrated demand response scheduling for commercial buildings across multiple time scales.

Images