Virtual power plant risk avoidance optimal energy and standby scheduling method considering demand response

Abstract

The invention discloses a virtual power plant risk avoidance optimal energy and standby scheduling method considering demand response. The technology aims at solving the defects of an existing virtual power plant scheduling method in the aspects of coping with uncertain factors such as renewable energy power generation, load demand changes and market price fluctuations, mining demand response potential, refining risk management and the like. By constructing a multi-stage stochastic optimization model covering a demand response plan and integrating a self-elastic and cross-elastic response type load model, the initiative of customers is accurately described, and conditional value-at-risk (CVaR) is adopted as a risk management tool, so that flexible adjustment and economic operation benefit maximization of a virtual power plant in an uncertain environment are realized. Through verification, the method can significantly improve the expected profit of the virtual power plant, effectively reduces the risk, and especially has a good effect when customers participate in load reduction and transfer at the same time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of virtual power plant participating in power grid dispatching, and particularly relates to a virtual power plant risk-avoiding optimal energy and reserve scheduling method considering demand response. TECHNICAL BACKGROUND

[0002] With the advancement of energy transformation and the large-scale access of distributed energy resources (DER), the operation and management of power systems are facing new challenges and opportunities. In this context, virtual power plant (VPP) emerges as an innovative energy management concept, which integrates distributed generation units, energy storage facilities, and adjustable loads, etc. resources together, and realizes the coordinated optimization scheduling of various energy resources through advanced communication, control and optimization technologies, to improve energy utilization efficiency, reduce operating costs, and enhance the stability and reliability of power systems.

[0003] Currently, the energy and reserve scheduling method of virtual power plant has attracted widespread attention, and numerous researchers and enterprises have invested a lot of effort in related research and practice. Among them, the existing methods closer to the present application technology mainly focus on deterministic scheduling and simple stochastic scheduling strategies. For example, some traditional virtual power plant scheduling methods only consider deterministic load demand and renewable energy generation, based on fixed price signals and system operation constraints, to develop day-ahead market trading plans and real-time operation strategies. This kind of method can realize the economic operation of virtual power plant to a certain extent, but when facing the intermittency and uncertainty of renewable energy generation, the volatility of load demand, and the frequent changes of electricity market price, it often exposes many shortcomings. First, they cannot accurately predict and quantify the impact of uncertainty factors on virtual power plant operation, resulting in reduced feasibility and reliability of scheduling plans. In actual operation, there may be insufficient or excessive reserve capacity, which not only affects the power supply quality of the power system, but also increases the operating cost and risk of virtual power plant. Secondly, this kind of method usually regards demand-side resources as passive loads, and fails to fully tap and utilize the potential of demand response to optimize the operation of virtual power plant. Demand response can effectively regulate load demand, alleviate the imbalance between supply and demand, and improve the flexibility and economy of the power system, but the existing methods fail to fully consider the enthusiasm and initiative of customers participating in demand response, as well as the complex impact of different demand response strategies on virtual power plant scheduling decisions, thereby limiting the optimal scheduling effect of virtual power plant in uncertain environment.

[0004] In summary, the existing virtual power plant energy and reserve scheduling method still has deficiencies in dealing with uncertain factors, tapping the potential of demand response, and fine risk management and efficient solution, etc., and it is difficult to meet the efficient, economic and reliable operation requirements of virtual power plant in modern power system. Therefore, the present application proposes a virtual power plant risk-averse optimal energy and reserve scheduling method considering demand response, aiming to overcome the defects of the prior art and provide a more effective and more adaptive solution for the optimization scheduling of virtual power plant in complex environment. SUMMARY

[0005] The present patent considers virtual power plant in smart grid, which is composed of integration of distributed energy resources, i.e. dispatchable distributed generation units, wind turbines, traditional energy storage facilities and responsive loads. The virtual power plant operator simultaneously dispatches energy and reserve resources to meet local loads and maximizes profits by trading energy in day-ahead market (DAM) and real-time market (RTM). The virtual power plant operator decides to trade energy with the wholesale market according to the dispatching decision, energy and reserve prices and the production of renewable energy.

[0006] The active participation of customers in demand response programs has a significant impact on the operator's decision. Each customer has multiple responsive loads, including transferable loads and curtailed loads, as well as some non-responsive loads. Under this assumption, customers can participate in price-based demand response programs by managing the power consumption of smart home appliances, thereby reducing electricity bills.

[0007] In the present patent, the uncertainties of renewable energy power, load demand, day-ahead market, real-time market and reserve market prices, as well as the uncertainty of calling reserve services are considered. After generating scenarios for each parameter, the generated scenario set is combined to construct a scenario tree. Since the number of generated scenarios directly affects the computational complexity of the optimization problem, it needs to be reduced to a small number of scenarios that are sufficient to represent uncertainty.

[0008] In the proposed strategy, the scheduling is divided into two stages. In the first stage, the virtual power plant (VPP) submits hourly energy and reserve bidding decisions for the next day to the day-ahead market (DAM) and the reserve market (SRM). In this stage, decisions are made without knowing the future market prices, load demand and renewable energy generation. This means that these decisions are made non-anticipatorily with respect to the considered scenarios. The variables of this stage optimize the utilization scheduling problem before the realization of uncertainty.

[0009] In the second stage, the VPP decides the real-time energy and reserve capacity exchanged with the RTM and makes real-time scheduling decisions for distributed generation (DG), energy storage system (ESS) and demand side (DS) resources at each time period. The decisions in this stage are made after the scenario realization, including the state of DG, the optimal output power of DG, the load after implementing demand response (DR) plan, the reserve capacity of DG and DS resource deployment and the load curtailment loss. Due to the existence of random variables, the decision strategy of the VPP has a risk condition. Therefore, the conditional value at risk (CVaR) is used as a risk management tool for the management of the optimization problem to capture the risk aversion behavior of the VPP operator under different conditions.

[0010] The application is implemented by the following technical scheme: a virtual power plant risk-averse optimal energy and reserve scheduling method considering demand response, comprising the following operation steps:

[0011] S1, constructing a responsive load model containing self-elasticity and cross-elasticity

[0012] The goal of the consumer is to maximize their benefit, i.e. the revenue obtained from the VPP minus the dissatisfaction cost due to changing energy use.

[0013] Based on the proposed model, customers participate in demand response programs by applying load curtailment (LC) and load shifting (LS) options, using curtaillable and shiftable loads.

[0014] The concepts of self-elasticity ( ) and cross-elasticity ( ) are used to model the sensitivity of curtaillable and shiftable loads to price, respectively. and represent the demand sensitivity to price at time t and h, respectively, and are represented by (1) and (2), respectively:

[0015]

[0016] In the formula, ρ j,t is the electricity price provided to load j at time t; ρ j,h is the electricity price provided to load j at time h; is the initial value of the electricity price provided to load j at time t; is the initial value of the electricity price provided to load j at time h; D j,t is the electricity demand of customer j at time t; is the initial value of the electricity demand of customer j at time t.

[0017] When customer j participates in a price-based demand response (DR) program, it adjusts its response load from (initial value) to (final value) to achieve maximum benefit.

[0018]

[0019] The benefit of customer j can be expressed as:

[0020]

[0021] where, is the final benefit of load j after applying the demand response program; is the final income of load j after applying the demand response program.

[0022] To maximize the utility function of the customer, the following formula needs to be verified:

[0023]

[0024] where, S(D j,t ) is the benefit of load j after applying the demand response program; B(D j,t ) is the income of load j after applying the demand response program.

[0025] Considering the linear relationship between hourly load and electricity price, when customer j participates in DR only through the load curtailment (LC) option, its utility B(D DR (t)) can be expressed as:

[0026]

[0027] where, is the initial income of load j after applying the demand response program.

[0028] Taking the derivative of (6) with respect to D j,t and substituting it into (5), we get:

[0029]

[0030] In addition, the utility of customer j when participating in DR only through the load shifting (LS) option can be expressed as:

[0031]

[0032] where, T is time.

[0033] Therefore, when customer j participates in the DR program with both LC and LS options, the economic model of its demand can be expressed by combining (7) and (8) as:

[0034]

[0035] where λ j is the potential of the demand response program implemented by customer j; N T is the time interval.

[0036] S2, method for simplifying a multi-uncertainty scenario tree

[0037] Virtual power plants (VPPs) face multiple uncertainties during their operation, including the power generation of renewable energy systems, the prices of day-ahead market (DAM), real-time market (RTM), and spinning reserve market (SRM), the load demand, and the calling of reserve service. In this study, the prediction errors of the VPP random variables are modeled by their associated probability density functions (PDFs). The associated PDFs are calculated based on the historical records of the parameters in the studied environment. Here, the prediction errors of the load demand and the prices of different markets are modeled using normal distributions, whose PDFs can be expressed as:

[0038]

[0039] where x represents the uncertain parameter, δ is the standard deviation parameter, μ is the mean of the uncertain parameter, which is equivalent to the predicted value of the associated variable. In addition, the Weibull distribution is used to model the wind speed uncertainty, as follows:

[0040]

[0041] where v, φ, and s are the wind speed, shape parameter, and scale parameter, respectively.

[0042] The uncertainty of the called reserve represents the deviation between the reserve actually deployed in response to the load and the amount of reserve called by the VPP. Considering the response ratio κ j,t is the reserve actually deployed by customer j in response to the load to the estimated reserve r j,t , which can be expressed as:

[0043]

[0044] where κ j,t is a random variable following a normal probability density function.

[0045] In this study, Monte Carlo simulation (MCS) method is adopted to generate scenarios by randomly sampling the probability distribution function of each random parameter. Firstly, 100 scenarios are generated for each mentioned random variable, then these generated scenarios are combined to obtain a scenario tree containing 1012 scenarios, which leads to a difficult-to-handle optimization problem. To deal with the problem caused by such a large number of scenarios, the K-means clustering method is adopted to reduce the number of scenarios to 200, thus reducing the computational burden.

[0046] The objective function of this problem aims to maximize the profit EP of the virtual power plant (VPP), which includes the profit item P H&N related to the on-demand decision and the profit item P W&S related to the standby decision, as well as the item CVaR of the CVaR tool. Therefore, it can be represented as:

[0047] Maximaze EP=[P H&N +P W&S +β×CVaR](13)

[0048] In the formula, CVaR is multiplied by a weight parameter β in the model, where β is used to balance the profit of the virtual power plant (VPP) and the risk of profit fluctuation. Risk-averse operators choose a larger β value to increase the risk weight, while risk-neutral operators tend to choose a higher risk to obtain higher profits, so the value of β is set close to zero; P H&N = ψ1- ψ2- ψ3, where ψ1 to ψ3 can be captured by the following equations:

[0049]

[0050]

[0051] In the formula, the first term of ψ1 represents the income of the virtual power plant from energy transactions with the main grid in the day-ahead market (DAM) and the sale of electricity to customers, N S and s is the corresponding scenario, π s is the occurrence probability of scenario s, is the total active power sold by the virtual power plant in the day-ahead market, is the day-ahead electricity sales price; the second term of ψ1 represents the income of providing backup services to the grid, N J is the load group, is the load reduction active power of customer j, ρ j,t is the electricity price provided to load j at time t, N G and i is the distributed power source number, is the spinning reserve backup electricity sales price, Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon i Pup, Pdown, Pnon, Pup, Pdown, Pnon i,t,s Pup, Pdown, Pnon, Pup, Pdown, Pnon i Pup, Pdown, Pnon, Pup, Pdown, Pnon i,t,s Pup, Pdown, Pnon, Pup, Pdown, Pnon i Pup, Pdown, Pnon, Pup, Pdown, Pnon i Pup, Pdown, Pnon, Pup, Pdown, Pnon i,t,s Pup, Pdown, Pnon, Pup, Pdown, Pnon i,t,s Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon Pup, Pdown, Pnon, Pup, Pdown, Pnon

[0052]

[0053]

[0054] wherein the term represents the cost of DG and DR reserve capacity deployed and the operating cost of battery energy storage system (BESS), which refers to its life cycle cost, and are the DG and DR reserve capacity offers, respectively, and are the DG up, down and non-reserve capacity, respectively, and are the DR up and down reserve capacity, respectively, N K and k are the energy storage system index, is the energy storage system k's energy capacity at time t, is the energy storage system k's charging capacity, is the energy capacity of the energy storage system k, Discharge loss for energy storage system k; Cost representing the mandatory load curtailment during the dispatch period, Loss of load value, Active power loss of load for customer j.

[0055] In addition, the CVaR is calculated as follows:

[0056]

[0057] where η s and ζ are the auxiliary variables and risk value for calculating CVaR, respectively, and a is the confidence level of the VPP. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 is a schematic diagram of the operation process of the virtual power plant risk-averse optimal energy and reserve scheduling method considering demand response of the present application;

[0059] Figure 2 is a single-line diagram of a 15-bus virtual power plant system;

[0060] Figure 3 is a virtual power plant load configuration diagram under different demand response conditions;

[0061] Figure 4 is Figure 3 a virtual power plant load risk-averse condition. DETAILED DESCRIPTION

[0062] In order to test the proposed scheduling method, a 15-node virtual power plant test system as shown in FIG. 1 is adopted. Figure 1 , Figure 2 The system includes three schedulable distributed power generation units, four wind turbines, three energy storage systems, and thirteen load nodes. Figure 3 The load curve is formed by collecting the electricity load of 2000 residential customers. Device A includes electrical devices with power not exceeding 200 watts, and device B includes air conditioning systems, fans, hair dryers, coolers, computers, exhaust fans, and other household appliances with power not exceeding 1000 watts. Considering that the daily load curve is divided into three different periods, i.e., the off-peak period (00:00-5:00), the non-peak period (5:00-10:00, 16:00-19:00, and 22:00-24:00), and the peak period (11:00-15:00 and 20:00-22:00). In addition, the expected values of the upward and downward regulation prices are assumed to be 1.1 and 0.9 times the day-ahead (DA) market price. The prices of upward and downward spinning reserve capacity are considered to be 15% of the DA electricity price.

[0063] To investigate the impact of DR in different VPP decision-making scenarios, the load curve characteristics, trading electricity with the main grid, expected profit, and CVaR index were studied for the following four cases:

[0064] Case 1: No DR plan implemented in VPP,

[0065] Case 2: Customers participate in DR only through LC options,

[0066] Case 3: Customers participate in DR only through LS options,

[0067] Case 4: Customers participate in DR through both LC and LS options.

[0068] The computational results of the two-stage stochastic optimization model with risk constraints described in the case study were obtained using CPLEX in GAMS software on a PC equipped with 4 GB of memory and an Intel Core i7@2.60 GHz processor.

[0069] Figure 2 The load profiles of the virtual power plant under the four cases are shown. As shown in the figure, in the second case, demand response measures are mainly applied based on the load curve options, resulting in a reduction in electricity demand during peak hours to reduce electricity bills, but no changes in other periods. However, in the third case, customers reduce electricity consumption during peak hours and shift part of their electricity consumption to other periods, especially during off-peak hours. It is important to note that the total daily electricity demand remains unchanged before and after load adjustment, but by changing the electricity consumption pattern, electricity costs can be reduced.

[0070] Figure 3 、 Figure 4 The relationship between the expected profit and CVaR of VPP and the risk aversion parameter β under different cases was compared. It can be seen that as β increases, the profit of VPP decreases, which is due to the occurrence of undesirable results in the worst-case scenario under risk neutrality. However, in case IV, customers participate in the demand response program using both curtailed and shifted loads, resulting in the highest profit and the lowest CVaR. In addition, as β increases, the expected profit of VPP decreases and the CVaR increases in all cases. However, at a lower β value (for the given case study, this level is lower than 1.6), the impact of risk aversion on profit and CVaR is negligible, and it is not effective to control the loss of expected profit. On the contrary, at a higher β value, more profit reduction is observed in all cases due to an increase in the number of unfavorable scenarios with negative profits.

[0071] To further investigate the proposed strategy in terms of computational efficiency and economic indicators, the results in different scenarios are compared with a stochastic model similar to the one described in [“A multi-time-scale economic scheduling strategy for virtual power plant based on deferrable loads aggregation and disaggregation,” IEEE Trans. Sustain. Energy, vol. 11, no. 3, pp. 1332–1346, Jul. 2020.], where the reserve reservation of DS resources is ignored. This model, hereafter referred to as the approximate model, is a modified version of our robust model, where the operation formula of the scheduling unit is replaced by the one proposed in the cited document. Table 1 shows the expected profit of the VPP, the total reserve cost, and the computational time of the optimal solution of the proposed model and the approximate model in different risk-aversion cases. In terms of computational effort, the proposed strategy takes less time to compute than the approximate model in all conditions. As shown in the table, the total reserve cost decreases when DS resources are used to provide spinning reserve service, especially in Case IV, where customers participate in both LC and LS options. Moreover, the participation of DR actors in spinning reserve service leads to an increase in the expected profit of the VPP. In contrast, for the approximate model, the reserve cost increases, resulting in a decrease in the expected profit. These results show that the proposed strategy is a more economic way of scheduling for the VPP.

[0072] Table 1 Comparison of results of the proposed method with other methods

[0073]

[0074]

[0075] A virtual power plant risk-averse optimal energy and reserve scheduling method considering demand response is proposed. By incorporating demand response schemes into the model, the impact of different demand response actor choices on virtual power plant decisions is also discussed. In addition, to address the uncertainties related to day-ahead and real-time market prices, renewable energy generation, load, and reserve service requests, the conditional value at risk (CVaR) is used for risk assessment. The model is verified on a 15-node virtual power plant system, and numerical results show that the use of different types of demand response measures can improve the profit of the virtual power plant. However, when customers participate in demand response using both load shedding (LS) and load curtailment (LC) mechanisms, the profit improvement effect is more significant.

Claims

1. A virtual power plant risk avoidance optimal energy and reserve scheduling method considering demand response, characterized in that, include: At least one processor; Memory connected to the at least one processor; Instructions stored in the memory, executable by the at least one processor, are used to perform the following steps: constructing a responsive load model incorporating self-resilience and cross-resilience to accurately characterize the initiative and dynamism of customer participation in demand response; generating scenario trees and reducing scenarios based on Monte Carlo simulation and K-means classification to effectively handle multidimensional uncertainties; employing Conditional Value at Risk (CVaR) as a risk management tool to construct a refined risk management framework; establishing a two-stage stochastic programming model, including first-stage decision-making in the day-ahead market and reserve market, and second-stage decision-making in the real-time market and distributed generation, energy storage, and demand-side resources; and solving the two-stage stochastic programming model using an optimization algorithm to obtain the optimal energy and reserve dispatch strategy for the virtual power plant.

2. The method according to claim 1, characterized in that, The responsive load model includes: customers participating in the demand response program by applying load reduction (LC) and load shifting (LS) options; self-resilience and cross-resilience used to model the price sensitivity of reduceable and shiftable loads; and a customer benefit model used to represent the benefits customers receive after participating in the demand response program.

3. The method according to claim 1, characterized in that, The multidimensional uncertainties include fluctuations in renewable energy power, changes in load demand, uncertainties in day-ahead market prices, real-time market prices and reserve market prices, as well as uncertainties in calling up reserve services.

4. The method according to claim 1, characterized in that, The method for calculating the conditional value at risk (CVaR) includes: determining the confidence level α of the virtual power plant; calculating the risk value for each scenario based on each scenario in the scenario tree; and calculating CVaR based on the risk value and scenario probability.

5. The method according to claim 1, characterized in that, The objective function of the two-stage stochastic programming model includes: maximizing the profit of the virtual power plant, including profit items related to spot decisions and profit items related to standby decisions; and balancing the profit of the virtual power plant with the risk of profit fluctuations, which is achieved by adjusting the risk aversion parameter β.

6. The method according to claim 1, characterized in that, The constraints of the two-stage stochastic programming model include: trading constraints between the virtual power plant and the main grid in the day-ahead market and the reserve market; operational constraints of distributed generation, energy storage and demand-side resources; load balancing constraints and reserve capacity constraints.

7. The method according to claim 1, characterized in that, The optimization algorithms include, but are not limited to, mixed integer programming, dynamic programming, or genetic algorithms.

8. The method according to claim 1, characterized in that, The virtual power plant includes dispatchable distributed generation units, wind turbines, energy storage facilities, and responsive loads.

9. The method according to claim 1, characterized in that, The scheduling strategy can adapt to different demand response plans and risk avoidance parameter settings to achieve flexible adjustment of the virtual power plant and maximize economic operating benefits.

10. The method according to claim 1, characterized in that, The system can provide decision support for virtual power plant operators under different risk preferences, so as to formulate scheduling strategies that are more in line with their own risk preferences.

Images