Virtual power plant double-layer extension planning method considering virtual energy storage

Abstract

The invention discloses a virtual power plant double-layer extension planning method considering virtual energy storage, and the method employs double-layer optimization to formulate an energy storage planning problem which comprises an upper layer and a lower layer, the upper layer is a planning problem and aims at minimizing the cost of energy storage configuration, and the lower layer is an operation problem and is used for optimizing the operation strategy of a VPP to achieve the maximization of operation benefits. The connection between the planning layer and the operation layer is realized by transmitting a capacity planning scheme of the ESS, in the upper-layer planning problem, the capacity of the ESS is a decision variable to obtain a planning scheme with the optimal cost, in the lower-layer operation problem, the output of scheduling resources is set as a decision variable, and the operation limitation of the scheduling resources is considered to realize the target of maximizing the operation benefit; the invention relates to the technical field of virtual energy storage. According to the virtual power plant double-layer extension planning method considering virtual energy storage, the advantage of maximizing the operation benefit while reducing the energy storage cost is realized, and the economical efficiency and the flexibility of the virtual power plant are improved.

Description

Technical Field

[0001] The present invention relates to the field of virtual energy storage technology, and in particular to a double-layer expansion planning method for a virtual power plant taking virtual energy storage into consideration. Background Art

[0002] Traditional physical energy storage systems, such as battery energy storage, although mature in technology and able to provide instant dispatch capabilities and stable services, their high installation and maintenance costs limit their large-scale promotion and application. In recent years, virtual energy storage systems, as a form of demand response, have gradually become an effective means to reduce energy storage costs and improve the flexibility of energy storage systems. Virtual energy storage incentivizes users to reduce or increase electricity consumption, thereby achieving similar functions to energy storage without relying on physical equipment, and has the advantages of low cost and high flexibility.

[0003] In a virtual power plant, reasonable energy storage system planning can effectively improve its economic and technical benefits. By combining physical energy storage and virtual energy storage systems, the virtual power plant can flexibly dispatch internal resources and balance electricity production and consumption. However, as the energy storage capacity increases, although the technical benefits will increase accordingly, the energy storage cost will also increase significantly. Therefore, how to strike a balance between cost and benefit has become a key issue in the energy storage planning of virtual power plants.

[0004] The present invention proposes a two-layer expansion planning method, which aims to optimize the operation strategy of the virtual power plant, reduce system costs, and improve economic benefits and operation flexibility by reasonably allocating the capacity of physical energy storage and virtual energy storage. Summary of the invention

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a two-layer expansion planning method for a virtual power plant considering virtual energy storage, which adopts two-layer optimization to formulate an energy storage planning problem, including an upper layer and a lower layer, wherein the upper layer is a planning problem, which aims to minimize the cost of configuring energy storage, and the lower layer is an operation problem, which is used to optimize the operation strategy of the VPP to maximize the operation benefit. The connection between the planning layer and the operation layer is achieved by transmitting the capacity planning scheme of the ESS. In the upper-layer planning problem, the capacity of the ESS is a decision variable to obtain the planning scheme with the best cost. In the lower-layer operation problem, the output of the scheduling resources is set as a decision variable, and their operation constraints are considered to achieve the goal of maximizing the operation benefit.

[0006] Preferably, the upper-level planning problem uses the capacity of physical energy storage and virtual energy storage as decision variables, and the objective function is to minimize the joint configuration cost of physical energy storage and virtual energy storage. The life cycle cost of physical energy storage includes capital, replacement, operation and recovery value and is related to the rated capacity and power level. The life cycle cost of virtual energy storage covers system design, operation and incentive or discount signal economic expenditures.

[0007] Preferably, the constraints of the upper-level planning problem include the energy storage system's rated capacity, power level, shortest time for full charge and discharge, and quantity being consistent with set values, the total energy storage capacity meeting the energy storage demand, and the total energy storage power meeting the scheduling demand.

[0008] Preferably, the lower-level operation problem constructs a VPP optimal operation model, and the objective function includes internal income and external income, the internal income includes energy supply, power generation cost, interruptible load compensation and energy storage discharge subsidy, and the external income includes energy transactions with distribution companies.

[0009] Preferably, the constraints of the lower layer operation problem include power balance, DG capacity, MT ramping power, IL scheduling, electric vehicle charging power and ESS operation constraints. The operation strategy is optimized based on the energy storage capacity and power boundary determined by the upper layer, and the upper and lower layer energy storage capacity and power ratings meet the interface constraints.

[0010] Preferably, it also includes a virtual energy storage system based on incentive or discount signals, which uses a utility function to consider customer preferences to determine the best incentives and discounts to adjust customer needs, designs a general VES model based on customer behavior, uses a quadratic function as a utility function to study consumer behavior, and the incentive rate or discount level is determined based on the VES discharge or charging level and customer elasticity. It is implemented through the network environment, and the operator determines the charging and discharging level according to the power situation and sends a signal to the customer to change consumption.

[0011] Preferably, a random programming method combining a roulette mechanism and simultaneous back substitution is used to replace wind speed and light intensity with probability density functions of Weibull and Beta probability distributions, generate scenarios based on the roulette mechanism, and reduce scenarios using simultaneous back substitution to simulate the uncertainty range.

[0012] Preferably, a hybrid metaheuristic algorithm combining a krill swarm algorithm and a gray wolf optimization is used to solve the bi-level programming problem. The krill swarm algorithm simulates the behavior of krill under the influence of neighborhood induction, foraging and physical diffusion to update the position to minimize the objective function. The gray wolf optimization imitates the gray wolf leadership and hunting mechanism to determine the α, β, and δ wolf-guided optimization. The hybrid metaheuristic algorithm updates the decision variables through the krill swarm algorithm and then is optimized by the gray wolf. It has the advantages of avoiding local optimality, accelerating convergence and improving solution quality.

[0013] The present invention provides a two-layer expansion planning method for a virtual power plant considering virtual energy storage.

[0014] Beneficial effects:

[0015] A two-layer expansion planning model for virtual power plants considering virtual energy storage is proposed. By optimizing energy storage configuration and operation strategies, the advantages of maximizing operating benefits while reducing energy storage costs are achieved, and the economy and flexibility of virtual power plants are improved. A random programming method combining roulette mechanism and synchronous back-substitution method is used to deal with the uncertainty of renewable energy power generation. This method can effectively generate and screen a small number of representative scenarios, reduce computational complexity, and maintain an accurate characterization of the uncertainty of renewable energy generation. The population effect and random diffusion mechanism of the krill swarm algorithm can ensure that the search process has good diversity and reduce the possibility of falling into the local optimum. The gray wolf optimization can quickly converge to the global optimal solution by gradually narrowing the search range around the optimal solution. Due to the combination of the search mechanisms of the two algorithms, the hybrid metaheuristic algorithm can better balance exploration and development and improve the accuracy and stability of the solution. DETAILED DESCRIPTION

[0016] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0017] Embodiment, the present invention provides a technical solution: a two-layer expansion planning method for a virtual power plant considering virtual energy storage, which adopts two-layer optimization to formulate an energy storage planning problem, including an upper layer and a lower layer. The upper layer is used as a planning problem, aiming to minimize the cost of configuring energy storage, and the lower layer is used as an operation problem, optimizing the operation strategy of the VPP to maximize the operation benefit. The connection between the planning layer and the operation layer is achieved by transmitting the capacity planning scheme of the ESS. In the upper-layer planning problem, the capacity of the ESS is a decision variable to obtain the planning scheme with the best cost. In the lower-layer operation problem, the output of the scheduling resources is set as a decision variable, and their operation constraints are considered to achieve the goal of maximizing the operation benefit.

[0018] The upper-level planning problem aims to minimize the cost of configuring energy storage:

[0019] Upper objective function:

[0020]

[0021] in The cost of jointly configuring physical energy storage and virtual energy storage is and are the life cycle costs of physical energy storage and VES, respectively.

[0022] Life cycle costs of physical energy storage:

[0023] Life cycle costs of physical energy storage systems Including capital cost Replacement costs Operating costs And recycling value And it varies with the rated capacity and power level of the physical energy storage system.

[0024]

[0025] The capital cost includes equipment cost and conversion device cost, which depends on the rated capacity of the physical energy storage system. and rated power Taking into account the high investment in capital costs and replacement costs, the life cycle costs are converted into average annual cash flows and multiplied by the payback factor R.

[0026]

[0027] Replacement costs arise from the need to replace the physical storage equipment at the end of its life cycle and depend on the rated capacity and the number of replacements required during the project life cycle.

[0028]

[0029] Where ceil(x) means the smallest integer not less than x. Indicates the number of physical energy storage devices that need to be replaced.

[0030] The operating costs are related to the equipment costs and the conversion unit costs:

[0031]

[0032] The recycling value represents the income obtained from recycling a physical energy storage system that has reached the end of its life cycle:

[0033]

[0034] Life cycle costs of physical energy storage systems Rated capacity involved and rated power are decision variables, while the other symbols represent the corresponding parameters.

[0035] Life cycle costs of virtual energy storage:

[0036] The life cycle costs of virtual energy storage mainly focus on the economic expenditures of system design, operation and incentive / rebate signals.

[0037]

[0038] Among them Cdep It is the initial development cost, which is the initial development cost required to develop incentive and discount signal algorithms and user response prediction models, involving algorithm design, user behavior data analysis, system integration, etc. and is the cost of incentives and discounts, which will be described in detail later. is the system operation and maintenance cost, including the daily operation, maintenance and update of the signal system, including the server resources and personnel costs required to process data, optimize incentives and discount strategies, and calculate and publish signals. unc is the scheduling uncertainty cost caused by user response. Due to the uncertainty of user response to incentives or discount signals, the system cannot accurately predict and control power consumption changes. It may be necessary to reserve more spare capacity or perform additional scheduling operations, thereby increasing the system operation cost. Assuming that the additional power consumption deviation caused by uncertainty is ΔU, and the additional cost that the system needs to pay to deal with uncertainty is λ, the consumer response uncertainty cost is:

[0039] c unc =λ·ΔU

[0040] The total energy storage capacity and power of the system is composed of the sum of physical energy storage and virtual energy storage:

[0041]

[0042] in and are the rated capacities of physical energy storage and virtual energy storage, and are the rated powers of physical energy storage and virtual energy storage respectively.

[0043] The constraints of the upper-level planning problem include the ESS allocation limit:

[0044]

[0045] These include the rated capacity and power level constraints of the energy storage system, the minimum time for the energy storage system to be fully charged or discharged, and the constraints on the number of energy storage systems being consistent with the set value. It is an integer representing the shortest time required for the energy storage system to fully charge or discharge at a given power level. It should be enough to meet the energy storage needs of the system, and the total energy storage power Should be sufficient to meet scheduling requirements, N ESS is the number of physical energy storage facilities, N VES is the amount of virtual energy storage equivalent to virtual energy storage, N max It is the maximum number of units that can be put into production.

[0046] Lower-level operational issues, aimed at maximizing operational benefits:

[0047] The lower layer constructs the optimal operation model of VPP and optimizes the scheduling strategy of VPP to maximize operating income. The lower layer objective function is:

[0048]

[0049] Among them, F In and F Ex They are VPP’s internal income and external income:

[0050] VPP internal revenue calculation is as follows:

[0051]

[0052] Where R In is the energy supply income, which is related to the electricity supply and retail price of the virtual power plant, C DG is the generation cost of distributed generation, which depends on the output and cost coefficient of distributed generation, C IL is the compensation cost of the interruptible load, which is related to the curtailment power and compensation coefficient of the interruptible load. It is the subsidy for the discharge of the energy storage system, which is determined by the discharge power of the energy storage system and the subsidy coefficient. is the power reduction of the lower interruptible load, is the output power of the lower micro-turbine, is the output power of the lower energy storage system, and They respectively indicate that the energy storage system is in the discharging and charging states. is an integer variable representing the discharge state of the energy storage system, is the charging load of the electric vehicle, and is a decision variable, while other symbols represent corresponding parameters. VPP external revenue comes from energy transactions with distribution companies. The balancing market adopts a dual pricing system similar to the European electricity market. External revenue is calculated as follows:

[0053]

[0054] in and They are the electricity sales and electricity purchase of VPP. and are decision variables, while the other symbols represent the corresponding parameters.

[0055] The lower level constraints include:

[0056] Power balance constraints:

[0057]

[0058] in is the active power of the lower distributed generation, is the output power of the lower micro-turbine, is the output power of the lower energy storage system, is the curtailment power of the lower interruptible load, β is the line loss rate of the VPP system, and in the power balance constraint, and is the decision variable.

[0059] DG capacity constraints:

[0060]

[0061] MT climbing power constraints:

[0062]

[0063] IL scheduling constraints:

[0064]

[0065] These include the maximum power reduction constraint of the IL, the maximum total duration constraint of the IL, and the number of reduction constraints of the IL. is an integer indicating the IL reduction status, When IL is in load shedding state, , IL is not scheduled.

[0066] The charging power of electric vehicles needs to meet the energy demand constraints:

[0067]

[0068] Operational constraints of ESS:

[0069]

[0070] Where E i,t is the energy stored in the ESS at bus i at time t, c i,t and d i,t are integers representing the charging and discharging states of the energy storage system, P i rat,lo 、E i,t , c i,t and d i,t is the decision variable, and other symbols represent the corresponding parameters.

[0071] There are also interface constraints between the upper-level planning problem and the lower-level operation problem. The operation optimization of the lower-level VPP is based on the energy storage capacity and power boundary determined by the upper-level. In order to obtain the optimal operation strategy of the VPP under the corresponding energy storage capacity and power boundary, the energy storage capacity and power rating of the upper and lower levels should meet the following constraints:

[0072]

[0073] A virtual energy storage system designed based on incentive / discount signals:

[0074] VES can exchange energy with the power grid in response to external signals, thereby entering the electricity market. Customers reduce (increase) their electricity consumption according to incentives (discounts), which is similar to the discharge (charge) function of energy storage devices. The present invention designs a new model to determine the best incentives and discounts to persuade customers to adjust their needs. The utility function is used to consider customer preferences from a socio-economic perspective and study the impact of incentives / discounts on customers.

[0075] In addition to electricity charges, electricity consumption is also related to the customer's personal preferences, which can be classified from different perspectives, specifically: scale (large, medium, small), industry (residential, industrial, commercial and agricultural), customer income level (high, medium, low) and customer social / cultural level (high, ordinary, low). The utility function is used to represent the value that customers may obtain from electricity consumption. For this purpose, the utility function U(D,a) of each electricity customer is defined, where D is the electricity consumption level and a is the risk aversion coefficient that characterizes the customer's risk aversion behavior. The customer welfare under any electricity price π can be expressed as follows:

[0076] W(D,π,a)=A·U(D,a)-π·D

[0077] Where π represents the electricity price, and the customer’s goal of consuming electricity is to maximize his welfare, so:

[0078]

[0079] Assume that the customer pays price π 0 Consume D 0 As the initial power requirement, the calibration factor can be determined as:

[0080]

[0081] The utility function is an increasing and concave function, which gradually saturates with the increase of consumption. There is an optimal electricity consumption level, which is related to the electricity price and the customer's risk aversion coefficient. At the same time, there is a maximum electricity consumption level, because rational customers will not consume electricity when the welfare is negative. Different types of functions can be used as utility functions. A general VES model based on customer behavior is designed. The model can adapt to different types of utility functions. The quadratic function is a commonly used form of utility function. It is selected to study the consumption behavior of electricity customers. The quadratic utility function can be expressed as follows:

[0082]

[0083] For this form of utility function, the welfare function, calibration coefficients and consumption function are as follows:

[0084] W(D,π,a)=-A·(1-a·D) 2 -π·D

[0085]

[0086] Price elasticity is a standardized measure that estimates the change in customer demand in response to a one percent change in electricity prices and can be expressed as:

[0087]

[0088] The price elasticity of electricity demand of the quadratic utility function can be expressed as:

[0089]

[0090] Price elasticity is related to the risk aversion coefficient and customer electricity demand, where customers have a lower propensity to reduce demand by increasing risk averse behavior.

[0091] Virtual energy storage discharge model:

[0092] Incentive signals are used to persuade customers to reduce electricity consumption and provide discharge functions. The impact of incentives on electricity demand and the VES discharge model are described as follows:

[0093] Before receiving the reward, the customer's benefits are as follows:

[0094]

[0095] Where D 0 and π 0 are the consumption level and electricity price before receiving incentives respectively. The welfare maximization is as follows:

[0096]

[0097] The second derivative of the utility function will be as follows:

[0098]

[0099] After the incentive, customer demand is determined by D 0 Change to D dch , the formula is as follows:

[0100] ΔD dch =(D 0 -D dch )

[0101] Where ΔD dch is the customer's reduced electricity demand, corresponding to the discharge level. The customer receives incentives based on the reduction in his electricity demand. Therefore, the incentive cost can be expressed as:

[0102] C dch =π inc ΔD dch

[0103] Among them C dch and π inc They are incentive cost and incentive rate. After receiving the incentive, the customer's benefits will change:

[0104]

[0105] In addition, the first-order derivative of the customer's utility after receiving the incentive is:

[0106]

[0107] The Taylor series expansion of the customer utility function can be expressed as follows:

[0108]

[0109] The customer's utility function is as follows:

[0110]

[0111] Its in D dch The first-order derivative of a point is as follows:

[0112]

[0113] The incentive rate can be expressed as:

[0114]

[0115] The incentive rate is related to price elasticity, electricity price and demand reduction (discharge) level. In order to minimize the total incentive cost paid to all customers, the following Lagrangian optimization problem needs to be solved:

[0116]

[0117] Subject to:

[0118]

[0119] in and are the total incentive cost and the discharge level of VES respectively. The above optimization problem can be solved as follows:

[0120]

[0121] Among them, and L are Lagrange multipliers, Lagrange function constraints and Lagrange function operators respectively, i and j represent customer type indexes, N is the number of customer types, and in order to minimize the total incentive cost level, all customers should receive the same incentive rate, which is expressed as:

[0122]

[0123] Based on the above formula, the demand reductions for different customers will be related to each other as follows:

[0124]

[0125]

[0126] The reduction in demand for each customer can be determined as follows:

[0127]

[0128] Therefore, the incentive rate is finally expressed as follows:

[0129]

[0130] The total incentive cost can be expressed as:

[0131]

[0132] Virtual energy storage charging model:

[0133] In addition to incentives, customer demand will also change with changes in electricity prices. Although electricity prices are not allowed to rise, electricity discounts can be provided. Providing discounts will encourage consumers to increase electricity consumption, which is similar to the energy storage charging function. Assume that the discount level is equal to π d (i)=(π 0 -π ch ), and considering a one percent change in customer demand in response to a one percent change in electricity prices, the increase in demand (charging level) will be as follows:

[0134]

[0135] D ch =(D 0 -ΔD ch )

[0136] where π d , π ch , ΔD ch and D ch They represent the discount price, the electricity price after the discount, the increase in demand due to the discount (charging level), and the electricity consumption after the discount. Although the same discount is applied to all customers, different customers will increase demand in different ways according to their elasticity. The relationship between the increase in demand for different customers is as follows:

[0137]

[0138] VES charge level will equal the sum of the increases in demand as follows:

[0139]

[0140] Applying the same approach as determining the optimal incentive in the virtual energy storage discharge model above, the increase in demand due to discounted prices and discount levels is expressed as follows:

[0141]

[0142] The optimal discount rate is expressed as follows:

[0143]

[0144] The rebate is paid per MW consumed, including the increased demand, so the total rebate cost can be expressed as:

[0145]

[0146] In summary, the incentive rate (discount level) is determined based on the discharge (charging) level of the VES and the elasticity of the customer. The network environment can be used to implement the proposed VES, and the operator determines the charging / discharging of the virtual energy storage system based on its power shortage or excess power. level, taking into account charging / discharging, incentives / discounts (π inc / π d ) signal will be generated and propagated to customers through the network environment. Each customer will reduce / increase his consumption (ΔD) according to his risk aversion behavior and the received signal. dch / ΔD ch ).

[0147] A stochastic programming method based on the roulette wheel mechanism (RWM) and the simultaneous backward method (SBM) is used to deal with the uncertainty of renewable energy output:

[0148] In the proposed method, the wind turbine and photovoltaic output power are replaced by an appropriate probability density function (PDF). The uncertainty of wind turbine and photovoltaic output mainly comes from the random volatility of wind speed and solar radiation. Wind speed and light intensity follow Weibull and Beta probability distributions respectively:

[0149]

[0150] Where c and k are the scale parameter and shape parameter of Weibull distribution respectively; α and β are the shape parameters of Beta distribution respectively.

[0151] Each PDF region is divided into several segments or probability levels, each probability level represents a specific prediction / modeling error related to the uncertain parameters. Many scenarios are generated based on RWM. In each scenario, a random number in the range [0,1] is generated for the wind turbine and photovoltaic output power. This random number will fall into a certain probability level in the relevant probability density function, which indicates the possibility of a certain error in the wind turbine and photovoltaic output power. This process is repeated to generate a large number of scenarios.

[0152] S m =[s m1 ,s m2 ,s m3 ,...,s mn ]

[0153] In order to reduce the scene set and lower the computational requirements, the present invention uses SBM to perform scene reduction. The method selects the most dissimilar and most likely scene by calculating the distance between different scenes. The following steps are used to reduce the scene:

[0154] 1) Set λ s Consider it as the initial scene set. At the same time, assume that R is the set of remaining scenes after the scene reduction process, which is initially empty. The distance between any two scenes is calculated as follows:

[0155]

[0156] 2) Determine each scene R e Minimum distance to other scenes:

[0157] D ml=minD mm' m,m'∈N S m≠m'

[0158] Where l represents the number of scenes closest to scene s.

[0159] 3) Multiply the probability of each scenario by the minimum distance between that scenario and other scenarios:

[0160] PD m'l =ps l ·D m'l m'∈N S

[0161] 4) From the initial scene set λ s Delete the dth scene with the lowest value in the following criteria:

[0162] PD d =minPD m m∈N S

[0163] λ=λ-{d} R=R+{d} ps l =ps l +ps d

[0164] 5) Repeat steps 1) to 4) until the desired number of scenes is reached.

[0165] The above scenario reduction process will eliminate similar scenarios and scenarios with low probability. Therefore, the remaining scenario set is diverse enough to maximize the range of uncertainty in the simulation problem.

[0166] The hybrid metaheuristic algorithm (HMA) combining the krill herd algorithm (KHA) and grey wolf optimization (GWO) is used to solve the bi-level programming problem:

[0167] KHA is a swarm intelligence optimization algorithm based on bionics, which is inspired by the behavior of krill in nature. Krill seek more favorable environments through collective movement, foraging behavior and physical diffusion to maintain the survival of the population. The KHA algorithm abstracts this behavior of krill into a mathematical model and applies it to complex optimization problems.

[0168] The goal of KHA is to minimize an objective function (such as cost, error, or energy loss) by simulating the behavior of krill groups, whose position updates are affected by three key factors: neighborhood-induced motion, foraging behavior, and physical diffusion.

[0169] 1) Neighborhood-induced motion N i

[0170] Neighborhood-induced movement is the movement of individuals in a krill population due to the influence of neighboring krill. The movement direction of krill is determined by local effects, target effects, and repulsion effects.

[0171] N new,i =α i ·N max +ω·N old,i

[0172] Where N new,i is the new movement speed of the i-th krill, N max is the maximum induced velocity, α i is the motion direction factor, which is determined by the local group density and the target group density, and ω is the inertia weight, which controls the contribution of previous motion.

[0173] α i =α local,i +α target,i

[0174] where α local,i is the local influence of neighboring krill on the current krill, α target,i It is the guidance of the target individual (usually the one with the best fitness) to the current krill.

[0175] Neighborhood-induced movement simulates the response of individuals in a krill population to pressure from neighboring populations. Krill will try to move closer to high-density populations and away from areas with lower density. This behavior ensures the collective movement of the population.

[0176] 2) Foraging behavior i

[0177] Foraging behavior refers to krill moving guided by food locations and previous foraging experience, a process that helps krill discover new resources and rely on known optimal food locations.

[0178] F i =V f β i +ω f ·F old,i

[0179] Where V f is the foraging speed, β i is the food attraction, ω f is the inertia weight of feeding behavior, controlling the influence of previous foraging experience.

[0180] β i =β food,i +β best,i

[0181] where β food,i is the current food position attractive, β best,i It is the best food location that krill has ever found during foraging. The behavior reflects that krill gradually moves to high-yield areas by tracking the best food source. It uses the group's memory and previous foraging experience to enable krill to effectively lock in resource-rich areas when searching.

[0182] 3) Physical diffusion i

[0183] The process of krill randomly exploring the environment. By introducing randomness, krill individuals can avoid falling into local optimality and maintain the ability to explore the entire search space.

[0184]

[0185] Where D max is the maximum diffusion speed, I is the current iteration number, and I max is the maximum number of iterations, δ is a random direction vector whose value is randomly generated between [-1,1].

[0186] Physical diffusion provides a certain randomness for krill individuals, helping them to jump randomly in the search space and avoid search stagnation caused by over-concentrated local search.

[0187] 4) Krill position update formula

[0188] Based on the above three movement processes, the shrimp position update can be described by the following formula:

[0189]

[0190] in The sum of neighbor-influenced movements, foraging movements, and physical dispersal.

[0191] Algorithm flow:

[0192] 1) Initialization: randomly generate the initial position of krill and set parameters (such as maximum speed, foraging speed, diffusion speed, etc.);

[0193] 2) Fitness evaluation: Evaluate the positional fitness of each krill based on the objective function (such as the cost function or error function of the target optimization problem);

[0194] 3) Update krill positions: Update the position of each krill based on the combined effects of neighborhood-induced movement, foraging behavior, and physical diffusion;

[0195] 4) Boundary processing: ensuring that krill do not exceed the defined search boundary and adjusting their position if necessary;

[0196] 5) Iterative update: Repeat the above steps until the stopping condition is met (such as reaching the maximum number of iterations or the objective function converges);

[0197] 6) Output result: Return the krill position with the best fitness, which is the optimal solution to the problem.

[0198] The advantage of KHA is that it can combine local and global search strategies and prevent falling into local optimal solutions through random diffusion. The algorithm has been successfully applied in many fields, including engineering optimization, neural network training and clustering problems. The KHA algorithm solves optimization problems by simulating the clustering behavior, foraging process and random diffusion of krill. It is an effective bionic optimization method, which is particularly suitable for handling multi-dimensional complex optimization tasks and can find a balance between global exploration and local search.

[0199] GWO not only has the advantages of simple structure and few parameters like other metaheuristic algorithms, but also has an adaptive adjustment factor, which can better balance the diversification of GWO compared with other algorithms.

[0200] Gray wolves live in groups with a strict social hierarchy, which is one of the inspirations for GWO. Gray wolf groups can be divided into 4 levels: α, β, δ and ω. From high to low, each wolf plays a specific role in the pack according to the level to which the wolf belongs. α, also known as the dominant wolf, is responsible for planning and deciding the hunting behavior of the group and represents the best solution obtained by the algorithm so far. β follows α in ranking and its duty is to assist α in decision-making. δ, ranked third in the social hierarchy, is responsible for reconnaissance, surveillance and guarding the wolf pack, acting as scouts, sentinels and guardians. Although ω wolves have the lowest status in the tribe, their presence is crucial to maintaining peace in the group. In GWO's mathematical model, the first 3 best-fit solutions in the population are considered α, β and δ wolves, and the rest are considered ω wolves.

[0201] GWO not only imitates the leadership ability of gray wolves in nature, but also imitates the hunting mechanism. The mathematical model is based on the behavior of surrounding prey. Its action mode can be expressed by the following formula:

[0202]

[0203] A=2·α·r1-a

[0204] C=2·r2

[0205] a=2-2·(t / Max_iteration)

[0206] Where a is the convergence factor, which decreases linearly from 2 to 0 as the number of iterations increases, r1 and r2 represent random vectors in [0,1], A and C are coefficient vectors, D represents the approximate distance between the current wolf position and the prey, and X t and X t+1 Represent the positions of the wolf at the tth and t+1th iterations respectively, represents the position of the prey at the tth iteration. In order to describe the hunting mechanism mathematically, GWO assumes that the higher-ranked wolves in the wolf pack are closer to the prey, that is, α, β, and δ occupy the most powerful positions in the entire wolf pack. In this way, the positions of the remaining wolves are determined by these three wolves. The calculation formula is as follows:

[0207]

[0208] Where C1, C2 and C3 are defined random coefficient vectors, A1, A2 and A3 are defined coefficient vectors, and Indicates the positions of α, β, and δ.

[0209] The process of the GWO algorithm is as follows:

[0210] 1) Randomly generate population X i (i=1,2,...,n);

[0211] 2) Initialize GWO parameters (Max_iteration, a, A and C);

[0212] 3) Calculate the fitness value and assign α, β and δ;

[0213] 4) When the number of iterations is less than the default number of iterations, the position of the current gray wolf is updated for each gray wolf. After all gray wolves are updated, the parameters a, A and C are updated.

[0214] 5) Correct the position of the gray wolf outside the boundary line and calculate the fitness value of the new position;

[0215] 6) Update α, β and δ and repeat the cycle at t=t+1;

[0216] 7) Return to the position of gray wolf α.

[0217] In response to the above-mentioned two-level planning problem of virtual power plants, this patent proposes to combine the KHA and GWO algorithms, called the hybrid metaheuristic algorithm HMA.

[0218] The core principle of HMA is to combine the two optimization algorithms, KHA and GWO. Since the KHA-GWO algorithm updates the decision variables in two different channels, it increases the number of decision variable update processes and achieves the optimal solution with fewer convergence iterations and lower calculation time, improving the convergence speed and accuracy of the optimization process. The combination of the two algorithms can be summarized as follows:

[0219] 1) KHA update phase: First, the decision variables are updated through the KHA algorithm;

[0220] 2) GWO update phase: The GWO algorithm is then used to further optimize the solution generated by the KHA algorithm.

[0221] By alternating between the two algorithms, HMA can explore the solution space more efficiently and reduce the possibility of falling into a local optimal solution.

[0222] HMA optimization process combining KHA and GWO:

[0223] 1) Determine the random values ​​of N (population size) decision variables according to the constraints;

[0224] 2) Calculate the dependent variable according to the constraints, find F (fitness function) for N decision variable values, and then calculate the optimal value of F;

[0225] 3) When the convergence point is reached, the program ends; otherwise, it goes to the next step, and when the maximum number of iterations is reached, the convergence point can be obtained;

[0226] 4) Update the decision variables according to the optimal F from the previous step, first by KHA and then by GWO. Go to step 2).

[0227] Advantages of HMA:

[0228] HMA combines the advantages of KHA and GWO and has the following significant advantages:

[0229] 1) Avoid local optimal traps: KHA’s population effect and random diffusion mechanism can ensure that the search process has good diversity and reduce the possibility of falling into local optimality.

[0230] 2) Accelerate convergence speed: GWO can quickly converge to the global optimal solution by gradually narrowing the search range around the optimal solution.

[0231] 3) Improve the quality of solutions: By combining the search mechanisms of the two algorithms, HMA can better balance exploration and exploitation and improve the accuracy and stability of the solutions.

[0232] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A two-layer expansion planning method for a virtual power plant considering virtual energy storage, which adopts a two-layer optimization to formulate the energy storage planning problem, characterized in that: It includes an upper layer and a lower layer. The upper layer is a planning problem, which aims to minimize the cost of configuring energy storage. The lower layer is an operation problem, which is used to optimize the operation strategy of VPP to maximize the operation benefit. The connection between the planning layer and the operation layer is achieved by transferring the capacity planning scheme of ESS. In the upper-layer planning problem, the capacity of ESS is the decision variable to obtain the planning scheme with the best cost. In the lower-layer operation problem, the output of the scheduling resource is set as the decision variable, and their operation constraints are considered to achieve the goal of maximizing the operation benefit.

2. A two-layer expansion planning method for a virtual power plant considering virtual energy storage according to claim 1, characterized in that: The upper-level planning problem uses the capacity of physical energy storage and virtual energy storage as decision variables, and the objective function is to minimize the joint configuration cost of physical energy storage and virtual energy storage. The life cycle cost of physical energy storage includes capital, replacement, operation and recovery value and is related to the rated capacity and power level. The life cycle cost of virtual energy storage covers system design, operation and incentive or discount signal economic expenditures.

3. A two-layer expansion planning method for a virtual power plant considering virtual energy storage according to claim 2, characterized in that: The constraints of the upper-level planning problem include the energy storage system's rated capacity, power level, minimum time for full charge and discharge, and the quantity that is consistent with the set value.

4. A two-layer expansion planning method for a virtual power plant considering virtual energy storage according to claim 3, characterized in that: The lower-level operation problem constructs the VPP optimal operation model, and the objective function includes internal income and external income, namely, internal energy supply income, power generation cost, interruptible load compensation cost, and energy storage discharge subsidy income.

5. A two-layer expansion planning method for a virtual power plant considering virtual energy storage according to claim 4, characterized in that: The constraints of the lower-level operation problem include power balance, DG capacity, MT ramping power, IL scheduling, EV charging power, and ESS operation constraints. The operation strategy is optimized based on the energy storage capacity and power boundary determined by the upper layer, and the energy storage capacity and power ratings of the upper and lower layers meet the interface constraints.

6. A two-layer expansion planning method for a virtual power plant considering virtual energy storage according to claim 5, characterized in that: It also includes a virtual energy storage system based on incentive or discount signals. The virtual energy storage system uses a utility function to consider customer preferences to determine the best incentives and discounts to adjust customer demand, designs a general VES model based on customer behavior, and uses a quadratic function as a utility function to study consumer behavior. The incentive rate or discount level is determined based on the VES discharge or charging level and customer elasticity. It is implemented through the network environment. The operator determines the charging and discharging level based on the power situation and sends signals to customers to change consumption.

7. A two-layer expansion planning method for a virtual power plant considering virtual energy storage according to claim 6, characterized in that: A stochastic programming method combining the roulette mechanism and the simultaneous back substitution method was used to replace wind speed and light intensity with probability density functions of Weibull and Beta probability distributions. Scenarios were generated based on the roulette mechanism and reduced using the simultaneous back substitution method to simulate the uncertainty range.

8. A two-layer expansion planning method for a virtual power plant considering virtual energy storage according to claim 7, characterized in that: A hybrid metaheuristic algorithm combining krill swarm algorithm and gray wolf optimization is adopted to solve the bi-level programming problem. The krill swarm algorithm simulates the behavior of krill under the influence of neighborhood induction, foraging and physical diffusion to update the position to minimize the objective function. The gray wolf optimization imitates the gray wolf leadership and hunting mechanism to determine the α, β, and δ wolf-guided optimization. The hybrid metaheuristic algorithm updates the decision variables through the krill swarm algorithm and then is optimized by the gray wolf. It has the advantages of avoiding local optimality, accelerating convergence and improving solution quality.

Images