Virtual power plant energy allocation method considering multi-type load demand response

Abstract

The method relates to the field of virtual power plant energy distribution, and relates to a virtual power plant energy distribution method considering multi-type load demand response. According to actual power plant energy distribution and operation conditions, a virtual multi-type load demand response mechanism is used to complete optimal scheme determination. The method comprises the following steps: 1) completing source and load prediction and analyzing uncertainty; 2) formulating an IES multi-time scale optimal scheduling scheme; 3) completing day-ahead scheduling model and intra-day scheduling model constraints; 4) completing virtual power plant constituent unit uncertainty output modeling; 5) constructing a virtual power plant day-ahead transaction optimization model based on CVaR; 6) completing a virtual power plant day-ahead transaction optimization model solving method; 7) completing a virtual power plant carbon transaction method based on a reward and punishment ladder type and a demand response strategy maximizing benefits; 8) obtaining optimal demand response on the electricity consumption side; and 9) testing and analyzing the built energy distribution model based on multi-type load demand response.

Description

TECHNICAL FIELD

[0001] The method relates to the field of virtual power plant energy distribution and relates to a virtual power plant energy distribution problem considering multi-type load demand response. BACKGROUND

[0002] Nowadays, energy is the basis for human survival and development, and is a fundamental strategic resource related to national security and stability and national economic development. Under the vigorous development of the energy internet, a large number of producers and consumers appear in the energy system, and the randomness and volatility of a large number of distributed resources increase the complexity and control difficulty of the power grid, which has a major impact on the safe, reliable and economic operation of the power grid. Researching a power plant energy distribution method for multi-type load demand response has become an important step in solving the contradictions between increasing energy demand and energy shortage, and between energy utilization and environmental protection.

[0003] There is no very definite definition of virtual power plant (VPP) at home and abroad. The virtual power plant technology uses advanced sensing and control technology to effectively aggregate and dispatch new energy power generation, energy storage and other distributed resources, to participate in the auxiliary service market to obtain benefits, at the same time, to provide flexibility for the power grid, to improve the safety level of the power grid, and to reduce the operation cost and investment cost of the power grid. It is an important module in the multi-energy flow comprehensive energy management system. From the perspective of China's energy strategy and the development path of efficient energy saving and clean development, the comprehensive energy efficiency index of the virtual power plant should be considered from the perspectives of economic benefit, environmental benefit and energy efficiency improvement, and the operation planning of the virtual power plant also needs to take into account the above factors.

[0004] By building multiple types of energy equipment in the energy consumption area, a coordinated and optimized operation mode of multiple energies is established, the area can be integrated and transformed into a virtual power plant with a certain scale. When facing demand response, the virtual power plant can exercise the functions of load aggregator, aiming to integrate scattered demand response resources, unified control, and participate in the power grid demand response as a whole. Reasonable planning of power plant energy distribution through virtual power plant is conducive to improving comprehensive energy efficiency and achieving long-term sustainable clean development goals. At the same time, it also requires scientific and reasonable planning in the planning process, and optimizes the allocation of resources in the virtual power plant from the perspective of improving comprehensive energy efficiency. Therefore, how to coordinate multiple goals that need to be met in virtual power plant planning and seek a more reasonable power plant energy distribution method has become a crucial problem faced by virtual power plants. SUMMARY

[0005] The technical problem solved by the present application is to provide a virtual power plant energy distribution method considering multi-type load demand response.

[0006] The technical solution adopted by the present application is:

[0007] The application is a virtual power plant energy distribution method considering multi-type load demand response, and the optimal scheme is determined by the actual power plant energy distribution and operation condition of the virtual multi-type load demand response mechanism. First, source and load prediction is completed, uncertainty analysis is carried out, IES multi-time scale optimal scheduling scheme is made, day-ahead scheduling model and intra-day scheduling model are completed. After obtaining the prediction result and the model, the uncertainty analysis of the virtual power plant composition unit is completed. The virtual power plant day-ahead transaction optimization model of CVaR is constructed, and the virtual power plant day-ahead transaction optimization model solving method is completed. On the basis of the above process, the virtual power plant carbon trading method based on the reward and punishment ladder type is further completed, the deterministic scheduling result and the related influence relationship are analyzed, the optimal demand response of the power consumption side is obtained, and model test analysis is carried out.

[0008] The application is a virtual power plant energy distribution method considering multi-type load demand response, and the optimal scheme is determined by the actual power plant energy distribution and operation condition of the virtual multi-type load demand response mechanism. First, source and load prediction is completed, uncertainty analysis is carried out, IES multi-time scale optimal scheduling scheme is made, day-ahead scheduling model and intra-day scheduling model are completed. After obtaining the prediction result and the model, the uncertainty analysis of the virtual power plant composition unit is completed. The virtual power plant day-ahead transaction optimization model of CVaR is constructed, and the virtual power plant day-ahead transaction optimization model solving method is completed. On the basis of the above process, the virtual power plant carbon trading method based on the reward and punishment ladder type is further completed, the deterministic scheduling result and the related influence relationship are analyzed, the optimal demand response of the power consumption side is obtained, and model test analysis is carried out.

[0009] Step 1, complete source and load prediction and analyze uncertainty, make IES multi-time scale optimal scheduling scheme, complete day-ahead scheduling model and intra-day scheduling model. Among them:

[0010] Step 1.1, complete source and load prediction and analyze uncertainty.

[0011] The source and load prediction plays a key role in the economic and safe operation strategy of the power system and in the power transaction, and the key to whether the supply and demand of the microgrid can be balanced and the operation cost can be reduced lies in the prediction accuracy, therefore, it is crucial to improve the prediction accuracy. In the paper, the LSTM source and load prediction model is adopted to successfully improve the prediction accuracy.

[0012] Step 1.1.1, LSTM prediction principle.

[0013] 1) basic recurrent neural network model;

[0014] o t =g(Vs t )

[0015] s t =f(Ux t +Ws t-1 )

[0016] In the formula: o t is the neuron output; g is the activation function of the output layer; V is the weight coefficient of the output layer; f is the activation function of the hidden layer; x t is the current output; U is the weight coefficient of the current input; s t-1 is the state of the previous time hidden layer; W is the weight coefficient of the previous time state as the current time input;

[0017] 2) Add a forget gate.

[0018] The forget gate is used to control the information that needs to be saved from the previous cell state to the current cell state. The formula is:

[0019] f t = σ(W f · [s t-1 , x t ] + b f )

[0020] where s t-1 is the previous cell state; x t is the current input; W f is the weight matrix of the forget gate; b f is the bias of the forget gate; σ is the sigmoid activation function of the forget gate, and the final f t is a value of [0, 1]. If f t = 0, it means forgetting all the previous cell state, otherwise if f t = 1, it means remembering all the previous cell state. Usually, the value of f t is (0, 1), only remembering the information that needs to be saved in the previous cell state.

[0021] 3) Add an input gate.

[0022] The input gate controls how much information of the current input is saved to the current cell state. The formula is:

[0023] i t = σ(W i · [s t-1 , x t ] + b i )

[0024] where s t-1 is the previous cell state; x t is the current input; W i is the weight matrix of the forget gate; b i is the bias of the forget gate; σ is the sigmoid activation function of the forget gate; the final i t is a value of [0, 1]. If i t = 0, it means forgetting all the current input, otherwise if i t = 1, it means remembering all the current input. Usually, the value of i t is (0, 1), only remembering the information that needs to be saved in the current input.

[0025] 4) Add an output gate.

[0026] The output gate controls how much information in the current cell state is saved to the current output, and the calculation formula is:

[0027] o t = σ(W o · [s t-1 , x t ] + b o )

[0028] s t = o t · tanh(c t )

[0029] In the formula, s t-1 is the cell state at the previous time; x t is the input at the current time; W O is the weight matrix of the forget gate; b o is the bias of the forget gate; σ is the sigmoid activation function of the output gate; o t is a value in [0, 1]; and finally, the output s t at the current time is obtained by integrating o t obtained by the output gate and the current cell.

[0030] 5) Increase the cell state.

[0031] Before the current cell state, the candidate value vector of the current cell state needs to be obtained first The calculation formula is as follows:

[0032]

[0033] In the formula, s t-1 is the cell state at the previous time, x t is the input at the current time, W c is the weight matrix of the forget gate, b c is the bias of the forget gate, tanh is the activation function of the forget gate, is a value in [0, 1]; and c t-1 is the cell state at the previous time, f t , i t , are values at the current time, that is, the long-term memory and the current memory can be integrated to obtain the current cell state c t .

[0034] Step 1.1.2, analyze uncertainty.

[0035] Based on the prediction error dataset, the source load prediction error uncertainty level is obtained by fitting the discrete distribution, as shown in Table 1, Figure 1The source load prediction error distribution and fitting are shown (in turn, photovoltaic, fan, electrical load, and thermal load).

[0036] Table 1 Source load prediction error uncertainty level

[0037]

[0038] Taking the total electrical load in summer as an example, based on the source load short-term prediction model, the load prediction error data set is obtained, and the distribution of the prediction error distribution is fitted as shown in Figure 2 .

[0039] Step 1.2, develop an IES multi-time scale optimization scheduling scheme.

[0040] The IES multi-time scale optimization scheduling flowchart is shown in Figure 3 The developed multi-time scale optimization scheduling strategy includes a day-ahead scheduling model scheme and an intra-day scheduling model scheme, wherein the day-ahead scheduling interval is 15 minutes, the day-ahead operation plan of each adjustable energy equipment is developed for 24 hours, the intra-day scheduling interval is 5 minutes, the day-ahead plan is tracked, and the impact of power fluctuation is reduced by rolling optimization every 1 hour, further improving the model accuracy.

[0041] Step 1.3, complete the day-ahead scheduling model and the intra-day scheduling model.

[0042] In the process of completing the day-ahead scheduling model, the economic dispatching model is used for the day-ahead scheduling of the integrated energy system, the objective function is the minimum expected operation cost under all combined scenarios, and the objective function of the intra-day scheduling model is the minimum deviation of the adjustable energy unit, so as to ensure that the model is more perfect.

[0043] Step 1.3.1, complete the day-ahead scheduling model.

[0044] The economic dispatching model is used for the day-ahead scheduling of the integrated energy system, and the objective function is the minimum expected operation cost under all combined scenarios. As shown in the following formula.

[0045]

[0046] In the formula, C IES represents the expected operation cost of the system scheduling intra-day; respectively represent the external purchase energy cost and the equipment energy unit operation and maintenance cost under each scenario s; T represents the total period number of the scheduling period, and ΔT is the scheduling time interval; M, β s respectively represent the number of scenarios and the scenario occurrence probability value; λ gas (t) respectively represent the natural gas consumption and price of each period; λ grid(t) represents the power and price of electricity purchased from the external power grid in each period; γ s (t), C GE_open Respectively represent the number of gas engine starts in each period and the startup cost each time; λ GE They represent the gas turbine output in each period and its operation and maintenance cost per unit output; λ PV Respectively represent the photovoltaic output in each period and the operation and maintenance costs per unit output; λ WT They represent the wind power output in each period and the operation and maintenance costs per unit output; λ GB Respectively represent the gas boiler output in each period and its operation and maintenance costs per unit output; λ AC Respectively represent the output of the chiller and warm water machine in each period and the operation and maintenance costs per unit output; λ EC Respectively represent the output of the electric refrigeration machine in each period and its operation and maintenance cost per unit output; ),λ Bat Respectively represent the battery energy storage output in each period and its unit output operation and maintenance costs; λ HS They represent the thermal energy storage output in each period and its operation and maintenance costs per unit output; λ CS Represents the cold storage output and its unit output operation and maintenance costs in each period; the supply of cold and hot energy is different in different seasons, so when optimizing scheduling for a specific season, only The operating costs of unused equipment can be set to zero.

[0047] Step 1.3.2, complete the intraday scheduling model.

[0048] The objective function of the intraday scheduling model is to minimize the deviation of the adjustable energy unit, as shown in the following formula:

[0049]

[0050] Where T and Cn represent the total number of time periods and the number of adjustable energy units in the scheduling cycle respectively. They represent the intraday output value and the day-ahead output value of the i-th adjustable energy unit at time t respectively.

[0051] Step 2: Complete the uncertainty output modeling of the virtual power plant component units, build a CVaR virtual power plant day-ahead trading optimization model, and complete the solution method for the virtual power plant day-ahead trading optimization model.

[0052] Step 2.1: Complete the uncertainty output modeling of the virtual power plant components.

[0053] The completion of the uncertainty output modeling of the virtual power plant component unit is achieved on the basis of completing the basic component unit modeling of the virtual power plant. The component unit modeling here takes the energy storage device as an example, and its operation process is as follows: Figure 4 As shown, it ensures that it is more in line with the actual situation and improves the accuracy of the model, among which the uncertainty output modeling of wind turbine units and photovoltaic unit output modeling are the most important.

[0054] Step 2.1.1: Modeling the uncertain output of wind turbine units.

[0055] The uncertainty of wind turbine output depends on the random characteristics of wind speed, which is described by Weibull distribution. The wind speed calculation model is:

[0056]

[0057] Where v is the wind speed, c is the scale parameter of the Weibull distribution, and k is the state parameter. Based on the calculation of the wind speed probability density in the above formula, the relationship between the output of the wind turbine and the real-time wind speed is:

[0058]

[0059] Where, is the output power of the fan at time t; C p is the wind energy utilization coefficient; ρ represents the air density; A w It is the vertical projection area of ​​the wind speed on the area swept by the blades of the unit; is the rated power of the unit; V in 、V rated and V out It is the cut-in, rated and cut-out wind speed of the wind turbine.

[0060] Step 2.1.2: Modeling the uncertainty output of photovoltaic units.

[0061] The output uncertainty of photovoltaic generator sets depends on the random characteristics of solar radiation intensity, which is described by Beta distribution. The solar radiation intensity model includes:

[0062]

[0063] Where r is the solar irradiance during period t; r max is the maximum solar irradiance during period t; α and β are the shape parameters of the Beta distribution, and their changes will lead to changes in the shape of the Beta distribution probability density curve. α and β can be calculated based on the mathematical expectation μ and variance δ of the solar radiation intensity during that period.

[0064]

[0065] where μ and δ are the mathematical expectation and variance of the solar radiation intensity.

[0066] Output model of photovoltaic power generation based on calculation of solar radiation intensity.

[0067]

[0068] where x PV is the conversion efficiency, p PV is the total area of the photovoltaic module, θ t is the solar radiation intensity at t.

[0069] Step 2.2, constructing a virtual power plant day-ahead transaction optimization model of CVaR.

[0070] This process combines the flexibility of the energy storage system "low charge high discharge", stores the excess output, thereby reducing the positive deviation of the virtual power plant bidding electricity, and when the bidding electricity produces negative deviation, uses the energy storage unit and interruptible load mechanism to sell electricity and transfer part of the load, reduces the electricity deviation penalty cost in the day-ahead market settlement. The virtual power plant participates in the day-ahead market transaction flow chart is shown in Figure 5 .

[0071] The objective function is expressed as:

[0072]

[0073] wherein:

[0074] C VPP = C ESS + C WPP + C PV + C DR + C MT

[0075]

[0076] wherein, and are the operation costs of the wind power and photovoltaic units at t period; and are the depreciation costs of the wind power and photovoltaic units; g' WPP · p WPP and g' PV · p PV are the deviation costs of the wind power and photovoltaic units; a MT , b MT and c MT are the cost parameters of the MT unit; is the price of the energy bought and sold by the energy storage at t moment; The amount of electricity purchased by the energy storage at t time for charging; The operation cost of the energy storage at t time.

[0077] Step 2.3, completing the virtual power plant day-ahead transaction optimization model solving method.

[0078] The model solving method has a great influence on the model precision and actual application, and it is crucial to find a suitable model solving method, and the ant colony algorithm is adopted to complete the virtual power plant day-ahead transaction optimization model solving, and the model precision is further improved.

[0079] (1) Ant colony algorithm state transition probability.

[0080] In the ant foraging process, the behavior of ants is affected by the pheromone concentration, and the path selection will also change accordingly, and the probability of ant a transferring from node i to node j is represented as:

[0081]

[0082] In the formula, τ ij (t) is the pheromone on the path of ant a transferring from node i to node j at t time; η ij (t) is the expected degree of selecting from node i to node j, the longer the path is, the cheaper the optimal solution is, that is, the smaller the expectation is; is the distance set that can be reached by ant a from node i to node j.

[0083] (2) Ant colony algorithm pheromone update.

[0084] When each ant reaches the food point, it will leave pheromone on the path it has walked, that is, the pheromone concentration of the path is improved, and the change of the pheromone on the path can be represented as:

[0085]

[0086] In the formula, τ'(a) is the pheromone concentration at the latest position of ant a; alpha1 is the volatilization coefficient of the original pheromone on the path; delta tau j (a) is the pheromone left by the ant representing the optimal path in this iteration; tau(alpha) is the pheromone of the ant of the optimal path after the last iteration.

[0087] Step 3, completing the virtual power plant carbon trading method based on the reward and punishment ladder type and the demand response strategy of maximizing benefits, and finally obtaining the optimal demand response of the electricity side and performing test analysis.

[0088] This step is realized on the basis of multi-park system, and the multi-park system established has three parks, namely industrial park, commercial park and residential park. Different parks are connected through various networks to realize mutual communication and mutual understanding. The multi-VPP-IES structure established in this project is shown in Figure 6 The VPP technology is used to aggregate the resources within each IES park, so that each park forms a close whole, and each park is regarded as an independent VPP. The system contains four networks, namely radial natural gas network, radial power grid, ring heat network and ring cold network. The cold and heat interaction between parks is realized through cold and heat pipelines.

[0089] Step 3.1, complete the carbon trading method of virtual power plant based on reward and punishment ladder type and the demand response strategy of maximizing benefit.

[0090] Step 3.1.1, complete the carbon trading method of virtual power plant based on reward and punishment ladder type.

[0091] The reward and punishment ladder type carbon trading mechanism is constructed, and the carbon emission rights are divided into multiple intervals, and the carbon trading price increases in steps with the carbon quota. When the carbon trading volume is negative, the enterprise can sell the excess carbon quota to obtain the reward. Therefore, the carbon trading cost of t period is:

[0092]

[0093] In the formula, k is the carbon trading base price of the day; l is the interval length, which is taken as l=2t here; θ is the price increment, which is taken as 0.25 here; ω is the reward coefficient; C co2 (t) is positive, which means that the carbon emission exceeds at this moment, and the quota needs to be purchased from the carbon trading market, C co2 (t) is negative, which means that the carbon emission quota is not used up at this moment, and can be sold to obtain income. B C is the income of participating in the carbon trading market, and the carbon trading volume is E s (t).

[0094] On the basis of the complete carbon market trading method process as shown in Figure 7 , the carbon credit income is obtained.

[0095] C Jm =C c *υ*σ

[0096] In the formula, C Jm is the income of selling carbon credits, υ is the carbon credit conversion coefficient, and σ is the carbon credit base price.

[0097] Step 3.1.2, complete the demand response strategy of maximizing benefit.

[0098] The benefit obtained by the power seller is represented as.

[0099]

[0100] π * optimal price; optimal demand at optimal price; electricity price vector.

[0101] Step 3.2, get optimal demand response of electricity side.

[0102] Optimal demand response of electricity side is the most important process, which directly affects the power plant energy distribution.

[0103] The specific optimal demand response expression formula is as follows.

[0104]

[0105] Where m represents different demand responses; total load after demand response effect at t period; initial electricity consumption; and respectively, the saturation upper limit load and basic load at t period; exponential elasticity coefficient from x period to y period; ΔH m,x electricity price change amount of x period.

[0106] Step 3.3, test and analyze the demand response model built.

[0107] The test analysis process sets three operation scenarios: each VPP obtains its maximum profit before individual optimization; each VPP cooperates and participates in large grid trading and direct trading between VPPs; and each VPP cooperates and only has multi-VPP direct trading.

[0108] The test model is as follows.

[0109]

[0110] In the formula: respectively, the electricity load before and after demand response; change amount of transferable electricity load; rebound load of reduced electricity load; reduced electricity load respectively, the rebound coefficient of reduced electricity load; respectively, the reduced electricity load at time t-1, t-2, t-3, and the gas load expression is the same, only the subscript is different. The same period user will also make vertical demand response according to the difference between different energy prices at that time, and get the replaceable electricity load change amount and replaceable gas load change amount Advantages of the present application compared with prior art

[0111] (1) The source load prediction and uncertainty analysis are completed, the IES multi-time scale optimization scheduling scheme is formulated, the day-ahead scheduling model and the intra-day scheduling model are completed, in the simulation process, the influence of source load uncertainty is reduced, and the adjustment pressure of intra-day scheduling is reduced, and the implementation of day-ahead scheduling plan is ensured.

[0112] (2) The uncertainty analysis of virtual power plant composition unit is completed. The CVaR virtual power plant day-ahead transaction optimization model is constructed, and finally the virtual power plant day-ahead transaction optimization model solving method is completed. The deviation between prediction and actual output is reduced, and compared with other methods, the test process of the present application is closer to the actual situation.

[0113] (3) The virtual power plant carbon trading method based on reward and punishment ladder type and the demand response strategy of maximizing benefit are completed, and the energy distribution model based on multi-type load demand response is tested and analyzed. The feasibility of the method is verified by analysis. It can improve the economy and safety of power plant operation, and effectively improve the operation efficiency of the power plant. The business management function of the operation and dispatching control platform can provide unified user service management, and assist the power plant to realize safe and unified business management. BRIEF DESCRIPTION OF DRAWINGS

[0114] Figure 1 Source load prediction error distribution and fitting condition.

[0115] Figure 2 Electric load prediction error distribution and fitting condition.

[0116] Figure 3 IES multi-time scale optimization scheduling flowchart.

[0117] Figure 4 Operation process of energy storage device.

[0118] Figure 5 Virtual power plant participating in day-ahead market transaction process.

[0119] Figure 6 Multi-VPP comprehensive energy system structure.

[0120] Figure 7 Participating carbon market transaction method process.

[0121] Figure 8 Gas turbine output comparison under 3 scenarios.

[0122] Figure 9 VPP traded electricity under scenario 1 and scenario 2.

[0123] Figure 10The VPP transaction electricity of each of scenarios 2 and 3.

[0124] Figure 11 The afternoon period resource response situation. DETAILED DESCRIPTION

[0125] (1) Based on LSTM, the wind power, photovoltaic output and electric and thermal load power are predicted, and compared with the actual values, the source and load prediction is completed, and the uncertainty analysis is carried out, so that the modeling process is closer to reality, the IES multi-time scale optimization scheduling scheme is formulated, the day-ahead scheduling model and the day-ahead scheduling model are completed, and see step 1 for details.

[0126] (2) The uncertainty analysis modeling of the virtual power plant composition unit is completed. The CVaR virtual power plant day-ahead transaction optimization model is constructed, and finally the virtual power plant day-ahead transaction optimization model solving method is completed, see step 2 for details.

[0127] (3) The virtual power plant carbon trading method based on reward and punishment ladder type and the demand response strategy to maximize the benefit are completed, the optimal demand response of the electricity side is obtained and tested and analyzed, see step 3 and the results obtained by the following test analysis for details.

[0128] The results obtained by the test analysis are as follows.

[0129] The VPP benefits under the three scenarios are shown in the following table, and the gas turbine output under the three scenarios is compared as Figure 8 .

[0130] The operation profit of each VPP under the three scenarios

[0131]

[0132] Figure 9 The total transaction electricity of VPP1 under scenarios 1 and 2 is VPP and the power grid and other VPPs. It can be seen that Figure 9 In the period of 11:00-15:00, VPP1 sells more electricity under scenario 2; in the period of 16:00-19:00, more electricity needs to be purchased due to insufficient wind power output. Since the direct transaction electricity price between VPPs is between the purchase and sale electricity price of the power grid, the transaction electricity under scenario 2 is obviously higher than that under scenario 1. In the calculation period of 24h, compared with scenario 1, the electricity purchase under scenario 2 increases by 38.35%, and the electricity sale increases by 15.21%.

[0133] The day-ahead transaction electricity of the three VPPs in scenarios 2 and 3 is as follows Figure 10As shown, the power greater than 0 is the sold electricity, less than 0 represents the purchased electricity from the outside, and in the period of wind power surplus in the VPP, the electricity trading is participated. In the period of 16:00-19:00, the wind power output of VPP2 is large, and a large amount of electricity is sold to the outside. Under scenario 3, since the electricity trading is limited between VPPs, the surplus wind power of VPP2 is consumed by the other two VPPs, and compared with scenario 2, VPP1 and VPP3 buy more electricity. It can be seen that the direct trading between VPPs reduces the burden of the power grid to a certain extent, and the joint operation of multiple VPPs provides a reference for the future construction of regional energy internet.

[0134] The virtual power plant response result is as follows.

[0135] The afternoon peak shaving response result is shown in the following table (where the amount of peak shaving and market price are known quantities), and the response of various resources in the virtual power plant is shown in Figure 11 .

[0136] Afternoon peak shaving response result table

[0137]

[0138] The response result of the virtual power plant in the afternoon period is compared with the response target amount obtained from the peak shaving market in the figure. It is concluded that in the afternoon period of the day, the peak shaving response of the virtual power plant is mainly completed by the air conditioning system and other adjustable loads. In each period of the afternoon, the internal resource response amount deviates slightly from the target value obtained in the market, and the energy storage device performs bidirectional compensation of charging and discharging, and the final result deviates by 0.032%, 0.056%, and 0.152% in each period, respectively, and the penalty cost at settlement is low.

Claims

1. A virtual power plant energy allocation method considering multi-type load demand response, comprising the following steps: 1) Complete source load forecast and analyze uncertainty; 2) Develop IES multi-timescale optimization scheduling scheme; 3) Complete the day-ahead scheduling model and intraday scheduling model constraints; 4) Complete the uncertainty output modeling of the virtual power plant components; 5) Construct a CVaR virtual power plant day-ahead trading optimization model; 6) Complete the solution method for the virtual power plant day-ahead trading optimization model; 7) Complete a virtual power plant carbon trading method based on a tiered reward and penalty model and a demand response strategy that maximizes benefits; 8) Obtaining the optimal demand response on the electricity consumption side; 9) Test and analyze the energy distribution model built based on multi-type load demand response; Step 7) completes the virtual power plant carbon trading method based on the reward and punishment ladder and the demand response strategy that maximizes benefits: Step 7.1: Virtual power plant carbon trading method based on reward and punishment ladder; A reward-and-penalty ladder-type carbon trading mechanism is constructed, which divides carbon emission rights into multiple intervals. The carbon trading price increases step by step with the carbon quota. When selling excess carbon emission quotas, a reward coefficient is introduced to incentivize the sale of excess carbon emission quotas to ensure that all unused carbon emission quotas can be sold. Therefore, the carbon trading cost in period t is: Where k is the carbon trading base price for the day; l is the interval length, which is taken as l = 2t here; θ is the price increase, which is taken as 0.25 here; ω is the reward coefficient; When it is positive, it means that the carbon emissions are in excess and it is necessary to purchase quotas from the carbon trading market. When it is negative, it means that the carbon emission quota has not been used up at this moment and can be sold to obtain income; B c To benefit from participating in the carbon trading market; Step 7.2: Demand response strategy to maximize benefits; The benefits obtained by the electricity seller are expressed as: π * For the best price; represents the optimal demand at the optimal price; represents the electricity price vector; Step 8) obtaining the optimal demand response on the electricity consumption side includes: The specific optimal demand response expression formula is as follows: Where m represents different demand responses; represents the total load after the demand response in period t; is the initial power consumption; and are the saturation upper limit load and base load in period t respectively; is the exponential elastic coefficient of period x to period y; ΔH m,x is the change in electricity price during period x; Step 9) Test and analyze the energy distribution model built based on multi-type load demand response: The test model is as follows: Where: are the electric loads before and after demand response, respectively; is the change in transferable electrical load; To reduce the rebound load of the electrical load; To reduce the electrical load; They are the rebound coefficients for reducing electrical load; The reduction in electric load at time t-1, t-2, and t-3 is respectively. The gas load is expressed in the same way, but with different subscripts. During the same period, users will also perform vertical demand response based on the difference between different energy prices at that time, and calculate the change in replaceable electric load under vertical demand response. and the change in the replaceable gas load 2. The virtual power plant energy allocation method considering multi-type load demand response according to claim 1 is characterized in that: The process of completing the source load prediction and analyzing the uncertainty in step 1) includes: Step 1.1: Establish an LSTM source-load prediction model; The Long Short-Term Memory (LSTM) neural network proposes a gate mechanism: forget gate, input gate, and output gate, and adds a cell state: In LSTM, the cell state is introduced; (1) LSTM principle; 1) Basic recurrent neural network model; o t =g(Vs t ) s t =f(Ux t +Ws t-1 ) Where: t is the neuron output; g is the activation function of the output layer; V is the weight coefficient of the output layer; f is the activation function of the hidden layer; x t is the current output; U is the weight coefficient of the current input; s t-1 is the state of the hidden layer at the previous moment; W is the weight coefficient of the state at the previous moment as the input at the current moment; 2) Add a forget gate; The forget gate is used to control the information that needs to be saved in the cell state of the previous moment to be saved in the current cell state; the calculation formula is: f t =σ(W f ·[s t-1 ,x t ]+b f ) Where s t-1 is the unit state at the previous moment; x t is the input at the current moment; W f is the weight matrix of the forget gate; b f is the bias of the forget gate; σ is the sigmoid activation function of the forget gate, and the final f t is a value in [0,1], if f t =0 means that the unit state at the previous moment is completely forgotten. On the contrary, if f t =1 means that all the unit states at the previous moment are remembered, usually f t The value of is (0,1), which only remembers the information that needs to be saved in the unit state at the previous moment; 3) Add input gate; The input gate controls how much information of the current input is saved to the current cell state. The formula is: I t =σ(W i ·[s t-1 ,x t ]+b i ) Where s t-1 is the unit state at the previous moment; x t is the input at the current moment; W i is the weight matrix of the forget gate; b i is the bias of the forget gate; σ is the sigmoid activation function of the forget gate; the final i t Is a value in [0,1], if i t =0 means forget all the input at this time. On the contrary, if i t =1 means all inputs at this time are memorized, usually i t The value of is (0,1), which only remembers the information that needs to be saved in the input at this time; 4) Add output gate; The output gate controls how much information in the current cell state is saved to the current output. The calculation formula is: the t =σ(W o ·[s t-1 ,x t ]+b o ) S t =o t *fishy(c) t ) Where s t-1 is the unit state at the previous moment; x t is the input at the current moment; W O is the weight matrix of the forget gate; b o is the bias of the forget gate; σ is the sigmoid activation function of the output gate; o t Is a value of [0,1]; the output s at the current moment t The output gate is o t After integration with the current unit; 5) Add unit state. Before the current unit state, you need to get the candidate value vector of the current unit state first. The calculation formula is as follows: Where s t-1 is the cell state at the previous moment, x t is the input at the current moment, W c is the weight matrix of the forget gate, b c is the bias of the forget gate, tanh is the activation function of the forget gate, is a value in [0,1]; where c t-1 is the unit state at the previous moment, f t 、i t 、 are all the values ​​at the current moment, which can integrate the long-term memory and the current memory to get the current unit state c t ; Step 1.2, uncertainty analysis; Based on the LSTM principle, wind power, photovoltaic output and electric, cooling and heating load power forecasts are completed and compared with their corresponding actual values; discrete distribution fitting is performed based on the prediction error data set to obtain the uncertainty level of the source-load prediction error.

3. The virtual power plant energy allocation method considering multi-type load demand response according to claim 1 is characterized in that: The formulation of the IES multi-timescale optimization scheduling scheme described in step 2) includes: (1) Formulate the objective function of the day-ahead scheduling model; The objective function is to minimize the expected daily operating cost under all combined scenarios, including the external energy purchase cost and the equipment energy unit operation and maintenance cost. The formula is as follows: Where C IES represents the expected operating cost of the system during the scheduling day; represent the external energy purchase cost and equipment energy unit operation and maintenance cost under each scenario s respectively; T represents the total number of time periods in the scheduling cycle ΔT is the scheduling time interval; M, β s Respectively represent the number of scenarios and the probability of scenario occurrence; λ gas (t) represents the natural gas consumption and price in each period; λ grid (t) represents the power and price of electricity purchased from the external power grid in each period; γ s (t), C GE_open Respectively represent the number of gas engine starts in each period and the startup cost each time; λ GE They represent the gas turbine output in each period and its operation and maintenance cost per unit output; λ PV Respectively represent the photovoltaic output in each period and the operation and maintenance costs per unit output; λ WT They represent the wind power output in each period and the operation and maintenance costs per unit output; λ GB Respectively represent the gas boiler output in each period and its operation and maintenance costs per unit output; λ AC Respectively represent the output of the chiller and warm water machine in each period and the operation and maintenance costs per unit output; λ EC Respectively represent the output of the electric refrigeration machine in each period and its operation and maintenance cost per unit output; λ Bat Respectively represent the battery energy storage output in each period and its unit output operation and maintenance costs; λ HS They represent the thermal energy storage output in each period and its operation and maintenance costs per unit output; λ CS Represents the cold storage output and its unit output operation and maintenance costs in each period; the supply of cold and hot energy is different in different seasons, so when optimizing scheduling for a specific season, only The operating costs of unused equipment can be set to zero; (2) Formulate the objective function of the intraday scheduling model; The objective function of the intraday scheduling model is to minimize the deviation of the adjustable energy unit, as shown in the following formula: In the formula, T and Cn represent the total number of time periods and the number of adjustable energy units in the scheduling cycle, respectively. They represent the intraday output value and the day-ahead output value of the i-th adjustable energy unit at time t respectively.

4. The virtual power plant energy allocation method considering multi-type load demand response according to claim 1 is characterized in that: The constraints of the day-ahead scheduling model and the intraday scheduling model in step 3) include: At any scheduling time, the system must meet the corresponding energy supply and demand balance constraints for cooling, heating, electricity, and gas. Different times have different corresponding energy balance constraints. The energy balance constraints for electricity, cooling / heating, waste heat, and gas are: Where, They represent the power purchased from the grid, wind power, and electric load power in each period under scenario s respectively; They represent the battery energy storage power, photovoltaic output value, and electric refrigerator power in each period under scenario s respectively; They represent the power of gas generators, productive energy storage batteries, water pumps, and electric vehicle charging power in each period under scenario s; They represent the gas grid power output value, gas boiler input power, and gas generator set input power in each period under scenario s respectively; They represent the output power of the electric refrigerator, the thermal energy storage power, and the unused thermal power of the gas generator set in each period under scenario s; They represent the cold energy storage power, gas boiler power, and gas generator set output heat power in each period under scenario s respectively; They represent the chiller / heater output power, cooling load power, heating load power, and thermal power input to the chiller / heater in each period under scenario s.

5. The virtual power plant energy allocation method considering multi-type load demand response according to claim 1 is characterized in that: The completion of the uncertainty output modeling of the virtual power plant component units in step 4) includes: (1) Uncertain output modeling of wind turbine units; The uncertainty of wind turbine output depends on the random characteristics of wind speed, which is described by Weibull distribution. The wind speed calculation model is: Where v is the wind speed, c is the scale parameter of the Weibull distribution, and k is the state parameter. Based on the calculation of the wind speed probability density in the above formula, the relationship between the output of the wind turbine and the real-time wind speed is: Where, is the output power of the fan at time t; C p is the wind energy utilization coefficient; ρ represents the air density; A w It is the vertical projection area of ​​the wind speed on the area swept by the blades of the unit; is the rated power of the unit; V in 、V rated and V out Cut-in, rated and cut-out wind speeds for wind turbines; (2) Uncertain output modeling of photovoltaic units; The uncertainty of photovoltaic generator output depends on the random characteristics of solar radiation intensity. Beta distribution is used to describe the solar radiation intensity model, which includes: Where r is the solar irradiance during period t; r max is the maximum solar irradiance during period t; α and β are the shape parameters of the Beta distribution. Their changes will lead to changes in the shape of the Beta distribution probability density curve. α and β can be calculated based on the mathematical expectation μ and variance δ of the solar radiation intensity during that period: Where μ and δ are the mathematical expectation and variance of solar radiation intensity; Output model of photovoltaic power generation based on calculation of solar radiation intensity: Where x PV is the conversion efficiency, ρ PV is the total area of ​​the photovoltaic module, θ t is the solar radiation intensity at time t.

6. The virtual power plant energy allocation method considering multi-type load demand response according to claim 1 is characterized in that: The construction of the CVaR virtual power plant day-ahead transaction optimization model in step 5) includes: Combined with the transaction demand of virtual power plants in the day-ahead market, its objective function is expressed as: in: C VPP =C ESS +C WPP +C PV +C DR +C MT Where, and is the operating cost of wind power and photovoltaic units in period t; and is the depreciation cost of wind power and photovoltaic units; g′ WPP ·p WPP and g′ PV ·p PV is the deviation cost of wind power and photovoltaic units; α MT , b MT and c MT is the cost parameter of the MT unit; The price of energy storage and electricity purchase and sale at time t; The amount of electricity purchased for charging the energy storage at time t; is the operating cost of energy storage at time t; During the model solution process, the market supply and demand balance and unit operation constraints are considered as follows: (1) Power supply and demand balance constraints in the day-ahead power market; Where D is the electricity demand, is the actual output of wind turbines in the day-ahead market, The actual output of the photovoltaic unit in the day-ahead market, The amount of electricity contributed by the energy storage unit to the day-ahead market; (2) Unit operating constraints; 1) Wind power output constraints; Where, is the upper limit of wind turbine output at time t; 2) Photovoltaic power generation constraints; Where, The upper limit of the photovoltaic unit output; 3) MT unit constraints; For MT units, the main considerations are power output and ramp constraints: Where, and They represent the lower and upper limits of MT output in period t respectively; and Respectively represent the up and down climbing power of the MT unit; 4) Energy storage unit constraints; AND min ≤E t ≤E max Where, Maximize charging efficiency for energy storage devices; is the maximum discharge efficiency of the energy storage device; δ s is the operating state of the energy storage system, that is, charging and discharging cannot be completed at the same time, E min With E max It is the minimum and maximum value of energy storage of the energy storage unit.

7. The virtual power plant energy allocation method considering multi-type load demand response according to claim 1 is characterized in that: The method for solving the virtual power plant day-ahead transaction optimization model described in step 6) is as follows: The ant colony algorithm is used to solve the virtual power plant day-ahead trading optimization model; (1) Ant colony algorithm state transition probability; During the foraging process of the ant colony, the behavior of the ants is affected by the pheromone concentration, and their path selection will also change accordingly. The probability of ant a moving from node i to node j is expressed as: Where, τ ij (t) is the pheromone on the path of ant a moving from node i to node j at time t; η ij (t) is the expected degree of selecting from node i to node j. The longer the path, the cheaper the optimal solution, that is, the smaller the expected degree. is the set of distances that ant a can reach from node i to node j; (2) Ant colony algorithm pheromone update; When each ant reaches the food point, it will leave pheromones on the path it has walked. That is, the pheromone concentration of this path increases. The change of pheromones on this path can be expressed as: Where τ'(a) is the pheromone concentration at the latest position of ant a; α1 is the volatility coefficient of the original pheromone on the path; Δτ j (a) is the pheromone left by the ant representing the optimal path in this iteration; τ(α) is the pheromone left by the ant belonging to the optimal path after the last iteration.

Images