Multiple risk management method based on conditional value at risk model under worst scenario

Abstract

A kind of multi-risk management method based on worst-case scenario conditional value at risk model, by setting the operation mode of virtual power plant in single electricity market, and after analyzing its bidding profit and constraint, the typical distribution scene of uncertain variable that will bring risk is constructed by using Monte Carlo simulation method and fast pre-generation elimination technology based on probability distance, then the conditional value at risk model under worst-case scenario is constructed, and the maximum conditional value at risk under worst-case scenario is taken as the goal, max-min problem is converted into semi-definite programming problem by using dual theory to solve, and the optimal bidding strategy of virtual power plant after management risk is solved.The present application can be used as a powerful tool for decision-making, provide theoretical support for virtual power plant to formulate scheme to improve the prediction accuracy of random variable, so as to minimize the market risk caused by prediction error, and make the transaction strategy closer to the optimal strategy under accurate situation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of virtual power plant control, and particularly relates to a multi-risk management method based on a worst-case scenario conditional value at risk model. BACKGROUND

[0002] When a virtual power plant carries out power transaction, it needs to face a series of uncertain factors, such as day-ahead market price fluctuation, output fluctuation of new energy distributed power (such as wind and light power units) caused by natural environmental factors, and load prediction accuracy problem. Only relying on the traditional risk management method can only manage one kind of risk, and cannot evaluate the loss of the virtual power plant, and causes the waste of adjustable load regulation capacity. SUMMARY

[0003] The present application proposes a multi-risk management method based on a worst-case scenario conditional value at risk model to solve the problems that the prior art does not consider the uncertainty factors of load prediction error and price fluctuation, is difficult to cope with complex and variable actual situations, and has low application efficiency in extreme conditions. The method first analyzes the bidding profit and constraint of the virtual power plant in the day-ahead main energy market, then adopts the Monte Carlo simulation method and the fast predecessor elimination technology based on the probability distance to construct typical distribution scenarios of uncertain variables that will bring risks, and finally constructs a worst-case scenario conditional value at risk model.

[0004] The present application is implemented by the following technical scheme:

[0005] The present application relates to a multi-risk management method based on a worst-case scenario conditional value at risk model, comprising:

[0006] Step 1) setting the operation mode of the virtual power plant in a single energy market, and analyzing the bidding profit and constraint thereof.

[0007] Step 2) adopting the Monte Carlo simulation method and the fast predecessor elimination technology based on the probability distance to construct typical distribution scenarios of uncertain variables that will bring risks.

[0008] Step 3) constructing a worst-case scenario conditional value at risk model, taking the maximum conditional value at risk in the worst-case scenario as the target, converting the max-min problem into a semi-definite programming problem by using the duality theory for solving, and obtaining the optimal bidding strategy of the virtual power plant after risk management. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 The present application is a flowchart;

[0010] Figure 2 The present application is a distributed photovoltaic output and distributed wind power output prediction curve;

[0011] Figure 3 For power load forecast curve;

[0012] Figure 4 For power load forecast 5000 scenarios to generate a graph;

[0013] Figure 5 For power load forecast 25 typical scenarios to generate a graph;

[0014] Figure 6 For different scenarios under virtual power plant day-ahead purchase strategy graph. DETAILED DESCRIPTION

[0015] As Figure 1 shown, the embodiment relates to a multi-risk management method based on the worst-case scenario under the conditional value at risk model, comprising:

[0016] Step 1, set the operation mode of virtual power plant in single electricity market, and analyze its bidding profit and constraints, specifically including:

[0017] 1.1 Consider the operation mode of virtual power plant in single electricity market, its income comes from two core market businesses of agent user power purchase and organization of flexible adjustable load to participate in demand response. In the power purchase and sale business, the price of virtual power plant selling electricity to users is based on the contract price, which is a fixed selling price, and the price of purchasing electricity from the grid follows the time-of-use price mechanism, which is the agent purchase price. In the demand response service, the system operator will implement certain economic incentive measures to optimize power supply and demand balance and achieve the goal of peak load shifting, that is, to compensate for the contribution of virtual power plant in demand response service. In order to encourage users to participate in demand response more actively, virtual power plant will also implement compensation strategy for the response amount of users participating in demand response, and set a certain compensation price. Based on the above setting of virtual power plant operation mode, the invention further discusses the profit sources and related constraints that virtual power plant needs to consider when bidding in day-ahead.

[0018] 1.2 According to the virtual power plant participating in the day-ahead main energy market bidding, the calculation of its net profit is based on total income and total cost, specifically: F LA = R LA - C LA , wherein: F LA is the net profit of virtual power plant participating in day-ahead main energy market, R LA is the total income of virtual power plant participating in day-ahead main energy market, C LA is the total cost of virtual power plant participating in day-ahead main energy market, wherein: the total income R LA of virtual power plant participating in day-ahead main energy market = R u + R idr , R uThe virtual power plant sells electricity to users, and the income is R idr The virtual power plant participates in demand response, and the income is R

[0019] The total cost C of the virtual power plant participating in the day-ahead main energy market LA = C E + C udr + C PV + C W + C ES , wherein: C E is the agent electricity purchasing cost of the virtual power plant, C udr is the compensation cost of the virtual power plant to the users participating in demand response, C PV is the light abandonment penalty cost of the distributed photovoltaic, C W is the wind abandonment penalty cost of the distributed wind power, and C ES is the charging and discharging cost of the energy storage device.

[0020] The virtual power plant sells electricity to users, and the income is R , wherein: e u is the contract electricity price of the virtual power plant to the users, P k,t is the actual power of the user load k at t after implementing demand response, N u is the number of users who entrust the virtual power plant to agent electricity purchasing.

[0021] The virtual power plant participates in demand response, and the income is R , wherein: e idr is the demand response compensation price paid by the system operator to the virtual power plant, ΔL idr,t is the demand response amount provided by the virtual power plant to the system operator at t.

[0022] The agent electricity purchasing cost of the virtual power plant is C , wherein: e E,t is the time-of-use electricity price of the day-ahead main energy market agent electricity purchasing at t, L t is the bidding amount of the virtual power plant in the day-ahead main energy at t.

[0023] The compensation cost of the virtual power plant to the users participating in demand response is C , wherein: e udr is the compensation price provided by the virtual power plant to the users participating in demand response; ΔP udr,k,t is the demand response amount of the user load k at t, N udr is the number of users participating in demand response.

[0024] The light abandonment penalty cost of the distributed photovoltaic is C , wherein: γ PV is the light abandonment penalty coefficient, P k (t) is the actual output power of distributed photovoltaic k at time t, P k (t) is the predicted output power of distributed photovoltaic k at time t, N PV N is the number of distributed photovoltaics.

[0025] The wind curtailment penalty cost of the distributed wind power where: γ W is the wind curtailment penalty coefficient, P k (t) is the actual output power of distributed wind power k at time t, P k (t) is the predicted output power of distributed wind power k at time t, N W N is the number of distributed wind power.

[0026] The charging and discharging cost of the energy storage device where: e ES is the charging and discharging cost price of the energy storage device, P ch,k,t and P dis,k,t are the charging and discharging power of the energy storage device k at time t, N ES N is the number of energy storage devices.

[0027] 1.3 Construct the constraint conditions of the virtual power plant when participating in the day-ahead main energy market bidding, including:

[0028] Power balance constraint: Energy storage device constraint: Distributed power supply constraint:

[0029] User load constraint: where: P ch,k,t and P dis,k,t are the charging and discharging power of the energy storage device k at time t, and are the charging and discharging states of the energy storage device k at time t, and and are both 0-1 variables, and are the maximum charging and discharging power of the energy storage device, η ch and η dis are the charging and discharging efficiency of the energy storage device, E b is the energy storage device battery capacity, SOC es,k,t is the state of charge of the energy storage device k at time t, SOC max and SOC min are the upper and lower limits of the state of charge of the energy storage device, and are the running upper and lower limits of the user load k at time t.

[0030] Step 2) Construct typical distribution scenarios of risky uncertain variables using Monte Carlo simulation and fast ancestral elimination technique based on probability distance, including:

[0031] 2.1 Obtain historical data and calculate the standard deviation s of historical data as an important indicator of data volatility. Then multiply the calculated standard deviation s with the standard normal distribution of generating random numbers to simulate volatility under different probabilities, and superimpose the volatility value on the predicted value of the original random variable y. Use Monte Carlo method to generate m equally probable scenarios, i.e. S pre (i) = D + sσ(i), i = 1, 2,..., m, where: S pre (i) is the i-th generated prediction scenario, D is the original scenario, and σ(i) is the generated random number.

[0032] 2.2 Use Manhattan distance to measure the difference between the generated m scenarios, i.e. where: d ij (i,j = 1, 2,..., m) is the Manhattan distance between the i-th scenario and the j-th scenario, ε is the total number of data in the scenario, X iu and X ju are the u-th data of the i-th scenario and the j-th scenario, respectively. Then calculate the Manhattan distance between each of the m scenarios, and get the corresponding distance matrix

[0033] 2.3 Calculate the sum of the Manhattan probability distance between each scenario and the remaining scenarios to get the probability distance matrix where: γ i (i = 1, 2,..., m) is the probability of the i-th generated scenario.

[0034] 2.4 According to the probability distance matrix, select the scenario with the smallest sum of probability distance with the remaining scenarios as the scenario to be reduced, and combine the two scenarios into a new scenario by taking the average value, i.e. S pre (k) is the newly generated scenario, γ k is its corresponding generation probability, S pre (g) and S pre (h) are the reduced scenarios, and γ g and γ h are their corresponding generation probabilities, respectively.

[0035] 2.5 Repeat steps 2.2) to 2.4) to continuously reduce scenarios, and finally get N reduced scenarios.

[0036] Step 3) Constructing the worst-case scenario condition value-at-risk model, and conducting risk assessment management, specifically including:

[0037] 3.1 When x and y are decision variables and random variables in the decision-making process, respectively, and the decision variable satisfies x∈X, X is the decision space, and the continuous probability density function of the random variable y is p(y), then for a given decision variable x, the cumulative distribution function of the loss function f(x, y) caused by y does not exceed the threshold α, Set the confidence level β∈(0, 1), then for a given decision variable x, the calculation expression of the value-at-risk VaR is V VaR-β (x) = min{α∈R:φ(x,α)≥β}, get its corresponding condition value-at-risk CvaR, Construct the auxiliary function Where: [f(x,y)-α] + As [f(x,y)-α] + = max{f(x,y)-α,0}, then the condition value-at-risk V CVaR-β (x) = minF β (x,α).

[0038] 3.2 The probability density function p(y) of the random variable y ∈ P, P is a distribution set of some known partial information, then for any fixed value x ∈ X, the worst-case scenario condition value-at-risk is Mix the distribution of the random variable, and construct its distribution set with a linear combination of several typical distributions of the random variable y known Where: N d is the total number of typical distributions of the random variable y, is the dth typical distribution of the random variable y, γ d is its corresponding weight coefficient; Construct the return function R(x, y) = -f(x, y), and assume that the number of random variables in the return function is W, and ignore the correlation between the random variables, i.e. the random variables will not affect each other, when each random variable obeys N ω typical distributions, then the total number of typical distributions is Under the construction method of the mixed distribution set, the worst-case scenario condition value-at-risk is

[0039] 3.3 Set the objective function as maximizing the worst-case scenario condition value-at-risk maxV WCVaR , convert the max-min problem to a semi-definite programming problem using duality theory to solve, and get the optimal bidding strategy of the virtual power plant after risk management.

[0040] Through specific experiments, the distribution of photovoltaic, distributed wind power output and load prediction data is set as shown in the figure. The day-ahead main energy market electricity price is predicted, and the specific price parameters are shown in Table 1.

[0041] Table 1 Price parameters

[0042]

[0043] In this embodiment, the confidence in the WCVaR model is set to 0.99, and it is assumed that the day-ahead electricity price prediction error, wind and light output prediction error and load prediction error all conform to the normal distribution. First, 5000 groups of scenarios are generated by using the Monte Carlo method, and then reduced to 25 scenarios by using the fast pre-generation elimination technology based on probability distance. Taking the power load prediction random variable as an example, the specific generation process of the corresponding scenario is shown in the figure.

[0044] According to the set embodiment, the bidding scenarios of the virtual power plant in the day-ahead main energy market are divided into:

[0045] Scenario one: the virtual power plant only agents the user to buy electricity.

[0046] Scenario two: the virtual power plant not only agents the user to buy electricity, but also organizes flexible adjustable load to participate in demand response.

[0047] The bidding strategy of the virtual power plant in the day-ahead main energy market under the corresponding scenario is obtained as shown in the figure.

[0048] And the corresponding WE and WCVaR values of the virtual power plant before and after participating in demand response are shown in Table 2

[0049] Table 2 WE and WCVaR of virtual power plant in different scenarios and corresponding ratio

[0050]

[0051] To further explore the sensitivity of the constructed WCVaR model, in this embodiment, the standard deviation of the day-ahead market electricity price prediction error, the distributed photovoltaic output prediction error, the distributed wind power output prediction error and the load prediction error are reduced by 1%, and the WCVaR value and WE value change in scenario two are shown in Table 3.

[0052] Table 3 WCVaR and WE value change of virtual power plant under different prediction errors

[0053]

[0054] According to the example, the WCVaR model under multiple risks constructed is analyzed, and its feasibility and effectiveness as a decision-making auxiliary tool are verified.

[0055] Compared with the prior art, the worst expected profit of the virtual power plant in the worst scenario is improved by participating in demand response, but the risk is increased, according to the WE and WCVaR of the virtual power plant in different scenarios and the corresponding ratio; according to the WCVaR and WE value change of the virtual power plant under different prediction errors, it can be known that improving the load prediction accuracy has the most significant influence on the risk and the income, and compared with this, the income increase and risk decrease brought by improving the wind power prediction and photovoltaic prediction accuracy are smaller.

[0056] In conclusion, the application can be used as a powerful tool for auxiliary decision-making, and can provide theoretical support for the virtual power plant to formulate a scheme for improving the prediction accuracy of random variables, so as to minimize the market risk caused by the prediction error, and make the transaction strategy closer to the optimal strategy in the accurate situation.

[0057] The above specific implementation can be adjusted in different ways by those skilled in the art without departing from the principles and purposes of the application, the protection scope of the application is subject to the claims and is not limited by the above specific implementation, and each implementation scheme within the scope is subject to the application.

Claims

1. A multi-risk management method based on a worst-case scenario conditional value-at-risk model, characterized in that, The virtual power plant is set in the operation mode in a single electricity market, and the bidding profit and constraint are analyzed, the typical distribution scenario of uncertain variable with risk is constructed by using Monte Carlo simulation method and fast pre-generation elimination technology based on probability distance, the condition risk value model under the worst scenario is constructed, and the maximum condition risk value under the worst scenario is taken as the target, the max-min problem is converted into semi-definite programming problem by using dual theory to solve, and the optimal bidding strategy of virtual power plant after management risk is solved; The risk management refers to: calculating the conditional risk value under the worst scenario according to the conditional risk value, and setting the objective function as maximizing the conditional risk value under the worst scenario The max-min problem is converted into a semi-definite programming problem by using the duality theory, and the optimal bidding strategy of the virtual power plant after the management risk is solved. The CVaR under the worst-case scenario is obtained by the probability density function of the random variable , , is a distribution set of certain known partial information, then for any fixed value , its CVaR under the worst-case scenario is ; mixing the distributions of random variables and using random variables The distribution set is constructed by linear combination of several typical distributions Wherein: is a random variable The total number of typical distributions, is a random variable The first typical distribution, is its corresponding weight coefficient; the profit function is constructed Let the number of random variables in the profit function be And ignore the correlation between each random variable, that is, the random variables will not affect each other, when each random variable obeys typical distribution, then the total number of typical distributions ; under the construction method of the mixed distribution set, the conditional value at risk under the worst scenario .

2. The method of claim 1, wherein the method is characterized by, The virtual power plant is set in the operation mode in a single electricity market, and the bidding profit and constraint are analyzed, and specifically includes: 1.1 The operation mode of virtual power plant in a single electricity market is considered, and the income is derived from two core market businesses of purchasing electricity by proxy users and organizing flexible adjustable load to participate in demand response, in the electricity purchase and sale business, the electricity price of virtual power plant sold to users is based on the contract electricity price, which is a fixed electricity sale price, and the electricity price purchased from the power grid follows the time-of-use electricity price mechanism, which is the proxy electricity purchase price, in the demand response service, the system operator will implement certain economic incentive measures to virtual power plant in order to optimize power supply and demand balance and realize the goal of peak clipping and valley filling, that is, the contribution of virtual power plant in demand response service is compensated, and virtual power plant also implements compensation strategy for the response amount of users participating in demand response in order to encourage users to participate in demand response more actively, and sets a certain compensation price, based on the setting of the above virtual power plant operation mode, the profit source and related constraints needed to be considered by virtual power plant in day-ahead bidding are further discussed; 1.2 The virtual power plant calculates its net profit based on total revenue and total cost when participating in the day-ahead main energy market, specifically: , wherein: is the net profit of the virtual power plant participating in the day-ahead main energy market, is the total revenue of the virtual power plant participating in the day-ahead main energy market, is the total cost of the virtual power plant participating in the day-ahead main energy market, wherein: the total revenue of the virtual power plant participating in the day-ahead main energy market , is the revenue of the virtual power plant selling electricity to users, is the revenue of the virtual power plant participating in demand response; 1.3 The constraint condition of virtual power plant participating in day-ahead main energy market bidding is constructed.

3. The method of claim 2, wherein the method is characterized by: The total cost of the virtual power plant participating in the day-ahead main energy market wherein: is the cost of purchasing electricity for the agent of the virtual power plant, is the compensation cost of the virtual power plant for participating in the demand response user, is the light abandonment penalty cost of the distributed photovoltaic, is the wind abandonment penalty cost of the distributed wind power, is the charge and discharge cost of the energy storage device; The virtual power plant sells electricity to users to obtain revenue wherein is the contract price of electricity sold by the virtual power plant to users, is the actual power of the user load at the time is the actual power of the user load at the time is the actual power of the user load at the time is the number of users who entrust the virtual power plant to purchase electricity on their behalf; The virtual power plant participates in demand response and the income Wherein: The demand response compensation price paid by the system operator to the virtual plant, The demand response compensation price paid by the system operator to the virtual plant, The demand response amount provided by the virtual plant to the system operator at the moment The agent electricity purchasing cost of the virtual power plant wherein: is the time-of-use electricity price of the day-ahead main energy market agent electricity purchasing at the time t, is the bidding quantity of the virtual power plant in the day-ahead main energy at the time t; The virtual power plant compensates the cost of the demand response user wherein: is the compensation price provided by the virtual power plant for the demand response user; is is the user load at the moment is the demand response amount of the user, is the number of demand response users. The aforementioned cost of curtailment penalty for distributed photovoltaic power. ,in: This is the penalty coefficient for discarded light. For a moment Distributed photovoltaic The actual output power For a moment Distributed photovoltaic Predicted output power The number of distributed photovoltaic systems; The distributed wind power curtailment penalty cost Wherein: is a wind power curtailment penalty coefficient, is a time The actual output power of the distributed wind power is a time The predicted output power of the distributed wind power is a time The predicted output power of the distributed wind power is the number of distributed wind power; The charge and discharge cost of the energy storage device Wherein: The charge and discharge cost price of the energy storage device, And Respectively, the time The charge and discharge power of the energy storage device At the moment, The number of energy storage devices.

4. The method for managing multiple risks based on the CVaR model under the worst scenario according to claim 2, wherein, The constraint conditions include: power balance constraint: ; energy storage device constraint: , distributed power supply constraint: , , user load constraint: , wherein: and are the charging and discharging power of the energy storage device at time , , and are the charging and discharging state of the energy storage device at time , , and are 0-1 variables, and are the maximum charging and discharging power of the energy storage device, and are the charging and discharging efficiency of the energy storage device, is the battery capacity of the energy storage device, is the state of charge of the energy storage device at time , , and are the upper and lower limits of the state of charge of the energy storage device, and are the running upper and lower limits of the user load at time , .

5. The method for managing multiple risks based on the CVaR model under the worst scenario according to claim 1, wherein, The typical distribution scenario is constructed by the following method, specifically including: 2.1 Obtain historical data and calculate the standard deviation of the historical data , as an important indicator of measuring data volatility, the calculated standard deviation is multiplied by the standard normal distribution of generating random numbers to simulate the volatility under different probabilities, and the volatility value is superimposed on the predicted value of the original random variable , and the Monte Carlo method generates equally probable scenarios 2.2 Differences between scenarios generated using Manhattan distance metric Differences between scenarios generated using Manhattan distance metric 2.3 Calculate the sum of the Manhattan probability distances between each scene and the remaining scenes to obtain the probability distance matrix. ,in: For the generated first The probability of each scenario; 2.4 According to the probability distance matrix, the scene with the minimum sum of probability distance to the remaining scenes is selected as the scene to be cut, and the two scenes are merged into a new scene by taking the average value, that is , is the newly generated scene, is the corresponding generation probability, and is the cut scene, and are the corresponding generation probabilities, respectively; 2.5 repeat steps 2.2) to 2.4) with ever decreasing scenes until a final pruned scene is obtained.

6. The method of claim 5, wherein the method is characterized by: The aforementioned An equally probable scenario refers to: ,in: For the generated first One predicted scenario, Original scene The generated random number.

7. The method of claim 5, wherein the method is characterized by: The difference refers to: Wherein: The Manhattan distance between the generated first scene and the first scene, The total number of data in the scene, And The first data of the first scene and the first data of the first scene, and then calculate the Manhattan distance between the two scenes, get the corresponding distance matrix .

8. The method for managing multiple risks based on the CVaR model under the worst scenario of claim 1, wherein, The conditional value at risk is obtained by and are the decision variable and the random variable in the decision-making process, respectively, and the decision variable satisfies , is the decision space, and let the continuous probability density function of the random variable be then for a given decision variable , the loss function caused by does not exceed the threshold of the cumulative distribution function ; set the confidence level , then for a given decision variable , the calculation expression of the value at risk VaR is , and the corresponding conditional value at risk CvaR is obtained , and the auxiliary function is constructed, where: is , then the conditional value at risk .

Images