CN116050800A - Distributed energy robust automatic scheduling method in multi-stage real-time auxiliary service market - Google Patents

Abstract

The invention discloses a distributed energy robust automatic scheduling method in a multi-stage real-time auxiliary service market, and relates to the field of robust scheduling strategies for automatic adjustment of virtual power plants. The invention comprises the following steps: formulating a deterministic multi-stage optimization model, defining optimization objectives, variables and limiting conditions; formulating a multi-stage distribution robust optimization model based on the deterministic multi-stage optimization model; under a specific fuzzy set, the multi-stage distribution The robust optimization model is transformed into a network structure to construct a dynamic programming model; the gradient descent algorithm is used to adjust the parameters in the value function of the convex optimization control strategy (COCP) in the dynamic programming model until the performance evaluation indicators converge, and the scheduling strategy is obtained. The present invention formulates a "multi-stage" distributed "robust" optimization model to characterize the scheduling strategy of the virtual power plant, respects the "causality constraints" of real-time operation, and considers the detailed physical model of the distribution network, including topology modeling and Exchange flow constraints.

Description

Robust automatic scheduling method for distributed energy in multi-stage real-time ancillary service market

Technical Field

The present invention relates to the field of robust scheduling strategies for automatic adjustment of virtual power plants, and more specifically to a method for robust automatic scheduling of distributed energy in a multi-stage real-time auxiliary service market.

Background Art

Traditional power system regulation services are achieved by dispatching large-scale synchronous coal-fired generators. However, in response to the goals of "carbon neutrality and carbon peak", the power system is also actively transforming and will no longer give priority to coal-fired generators in the future. On the other hand, with the popularization of a large number of distributed energy sources in the distribution network, they have become an effective and environmentally friendly means of real-time regulation of the power system. These distributed energy sources include renewable energy generators, energy storage systems, and controllable loads that can respond to demand. They are smaller in scale and more flexible in regulation. However, it is very difficult for system dispatchers to directly dispatch these distributed energy sources. First, they are numerous and scattered in the distribution system, requiring too many decision variables to be considered and strict and complex physical constraints to be met. Second, the output power of renewable energy generators is uncertain and intermittent, which may cause reliability problems in the power system.

The above problems gave rise to virtual power plants. First, virtual power plants can serve as a bridge between system scheduling and distributed energy, reducing the burden on power system dispatchers. It receives system scheduling instructions in real time, and centrally controls all distributed energy within its jurisdiction to jointly complete scheduling tasks. Secondly, virtual power plants effectively and robustly optimize distributed energy scheduling strategies for various real-time uncertainties to prevent reliability problems in the power system. These uncertainties include the regulation service requests of the upper-level scheduling system and the actual power generation of all renewable energy generators within its jurisdiction. How to design such a scheduling strategy is the core technical problem to be solved by the present invention.

Summary of the invention

In view of this, the present invention provides a method for robust automatic scheduling of distributed energy in a multi-stage real-time ancillary service market to aggregate distributed energy within the jurisdiction and provide real-time power system regulation services.

In order to achieve the above object, the present invention adopts the following technical solution:

A method for robust automatic scheduling of distributed energy in a multi-stage real-time ancillary service market comprises the following steps:

Develop a deterministic multi-stage optimization model, define optimization objectives, variables, and constraints;

Based on the deterministic multi-stage optimization model, a multi-stage distributed robust optimization model is formulated;

Under certain fuzzy sets, the multi-stage distributed robust optimization model is transformed into a mesh structure and a dynamic programming model is constructed.

Using the gradient descent algorithm, the parameters in the convex optimization control policy (COCP) value function generated by the dynamic programming model are adjusted until the performance evaluation index converges and the scheduling strategy is obtained.

Optionally, the calculation formula of the multi-stage distributed robust optimization model is as follows:

为决策变量的解空间。

表示数学期望，F(·)表示成本函数。Among them, G is the decision variable of scheduling, ζ is the uncertain variable, Q is the distribution of the uncertain variable,

is the solution space of decision variables.

represents the mathematical expectation, and F(·) represents the cost function.

Optionally, the dynamic programming model is as follows:

为决策变量的解空间。公式中的F_t(·)表示当前阶段成本函数,

表示未来阶段的价值函数，β表示状态变量。The subscript t represents the decision of the tth stage, V _t (·) represents the optimization target of the tth stage. G is the decision variable of the scheduling, ζ is the uncertain variable, Q is the distribution of the uncertain variable,

is the solution space of decision variables. F _t (·) in the formula represents the cost function of the current stage,

represents the value function of the future stage, and β represents the state variable.

Optionally, the distribution of uncertain variables in the dynamic programming model is as follows:

_ut and l _t are the upper and lower bounds of the uncertainty variables, respectively, and δ is the Dirac function; if μ _t is the average value of the uncertainty variables, the value of the weight α _t is ( _ut -μ _t )/( _ut -l _t ). Therefore, in actual operation, the information that needs to be known is the upper and lower bounds and the mean value of the uncertainty variables (the output of renewable energy units and the regulation service request of the upper system), so as to determine the most robust distribution.

Alternatively, a single-stage parameterized convex optimization control policy (COCP) is formulated as:

Optionally, the value function in COCP is expressed as a convex quadratic form, namely:

和

统称其为COCP参数，表示为θ_t+1。The parameters are

and

They are collectively referred to as COCP parameters and are denoted as θ _t+1 .

Optionally, the trained COCP parameters (θ={θ _t , t=1,…,T}) are directly introduced into the dynamic programming model, and single-stage deterministic optimization is performed according to the true value of the uncertainty variable at each stage to obtain the scheduling strategy for each stage.

It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for robust automatic scheduling of distributed energy in a multi-stage real-time auxiliary service market, which has the following beneficial effects:

1. A “multi-stage” distributed “robust” optimization model is developed to characterize the dispatch strategy of the virtual power plant, which respects the “causality constraints” of real-time operation. In addition, a detailed physical model of the distribution system is considered, including topology modeling and AC power flow constraints.

2. In order to make the scheduling and solving operation simpler and more effective, the multi-stage distributed robust optimization model is converted into an equivalent dynamic programming model. This conversion method is the innovation of the present invention. In order to further simplify the algorithm, the value function is approximated as a convex quadratic form, reducing the scale and computational complexity of the optimization model, and obtaining a real-time single-stage convex optimization control strategy (COCP).

3. An automatic algorithm is used to adjust the parameters in COCP. For the training algorithm, this approach guarantees better solution quality and less convergence time than the traditional fitting mechanism. For real-time operation, the proposed strategy outperforms other existing stochastic programming methods in terms of robustness, efficiency, and computational speed.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

Fig. 1 is a schematic diagram of the process of the present invention;

FIG2 is a feeder topology model of a virtual power plant according to the present invention and a diagram showing specific requirements for regulation services;

Figure 3 is a graph showing the relationship between the optimal values (conservativeness) of the multi-stage model, the network model, and the dynamic programming model;

FIG4 is a performance evaluation index calculation diagram of the approximate dynamic programming model of the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in 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 creative work are within the scope of protection of the present invention.

The embodiment of the present invention discloses a method for robust automatic scheduling of distributed energy in a multi-stage real-time auxiliary service market, as shown in FIG1 , comprising the following steps:

S1: Develop a deterministic multi-stage optimization model and define the optimization objectives, variables, and constraints;

S2: Based on the deterministic multi-stage optimization model, a multi-stage distributed robust optimization model is formulated;

S3: Under specific fuzzy sets, the multi-stage distributed robust optimization model is converted into a mesh structure and a dynamic programming model is constructed;

S4: Using the gradient descent algorithm, adjust the parameters in the convex optimization control policy (COCP) value function in the dynamic programming model until the performance evaluation index converges and obtains the scheduling strategy.

Furthermore, in S1, a deterministic multi-stage optimization model is formulated. In this model, the influence of uncertainty factors is not considered, and the purpose is to define the optimization objectives, variables and constraints. The specific demand for regulation services is that the virtual power plant needs to follow the system scheduling requirements and feed a certain amount of electricity through the connection node between the distribution system feeder and the high-voltage transmission system, as shown in Figure 2. It is assumed that the system pays the virtual power plant for providing regulation services has been agreed in the long-term contract. In the specific real-time scheduling, this part can be regarded as a "sunk cost". However, if the virtual power plant fails to contain the error of the power provided within a certain limit, it will be punished for shortage or surplus. Therefore, the optimization target is the power generation cost of distributed energy and the system's penalty for the output power deviation of the virtual power plant. The optimization decision variables are the output size of each distributed energy source, including the output of distributed small heat sources and new energy generators, the charging and discharging power of the energy storage system, and the controllable load demand response power. The constraints of the optimization problem are the AC power flow constraints of the distribution system and the output limits of distributed energy sources.

为决策变量的解空间：In S2, based on this deterministic multi-stage optimization model, a multi-stage distributed robust optimization model is formulated. The uncertainty considered in this model comes from the regulation service power required by the system and the output power of renewable distributed energy. Among them, distributed robust optimization is a compromise optimization method between robust optimization and stochastic programming. It is more robust than stochastic programming and has a certain degree of conservatism compared to robust optimization. Its purpose is to optimize the total cost under the "worst distribution" condition in the distributed fuzzy set of uncertain variables. In addition, the multi-stage distributed robust optimization model meets the "causal constraints" of real-time decision-making, that is, the decision at each stage can only be based on the known true values of uncertain factors and past decision results. The formula is as follows: G is the decision variable for scheduling, ζ is the uncertain variable, Q is the distribution of the uncertain variable,

is the solution space of decision variables:

In S3, the multi-stage distributed robust optimization model defined above can be converted into a mesh model, in which problems at different stages are nested together, following the idea of dynamic programming. On the other hand, since dynamic programming naturally satisfies the "causal restriction", by establishing the relationship between the multi-stage model, mesh model, and dynamic programming model, the optimal value relationship of the three is shown in Figure 3, and the size of the optimal value also reflects their degree of conservatism. The mesh model is equivalent to the dynamic programming model and is more robust (conservative) than the original model. By assuming that the distribution of uncertain variables belongs to a specific fuzzy set, the multi-stage robust optimization can be converted into an equivalent dynamic programming problem.

The specific assumption is: Assume that we only know the upper and lower bounds and the average value of the distribution of the uncertainty variable at each stage, and no more information. Under this assumption, we can directly get the "worst distribution" under the fuzzy set, and Q ^* is the worst distribution under the fuzzy set. In this way, we can write the specific value function in the simplified dynamic programming model.

Among them, _ut and _l are the upper and lower bounds of the uncertainty variable, _μt is the average value, and δ is the Dirac function.

Therefore, the problem is transformed into a stage-independent, risk-neutral and "worst distribution" stochastic programming, and its dynamic programming model is as follows.

Among them, the value function is expressed as the following formula:

The goal of each step is argmin V _t (β _t-1 , G _t-1 , ζ _[t] ).

Furthermore, by approximating the value function as a convex quadratic form, namely:

和

统称其为θ_t+1。The parameters are

and

They are collectively referred to as θ _t+1 .

The problem solving is then transformed into deriving a single-stage parameterized convex optimization control policy (COCP) at each time stage, which can be characterized as:

表示决策变量，β表示状态变量。具体而言，每个阶段的输出作为下一个阶段的输入，最终性能等于每个阶段成本总和的期望值。由于期望值往往不可直接计算，通常由蒙特卡洛方法生成其预估值。利用梯度下降算法，训练COCP在每个阶段的参数，算法流程下：In S4, based on the gradient descent algorithm, the parameters of the value function in COCP are automatically adjusted, that is, θ = {θ _t , t = 1, …, T}. The calculation of the overall performance evaluation index of COCP is shown in Figure 4, where in the tth stage, θ _t represents the parameters of COCP, ζ _t represents the uncertainty variable,

Denotes the decision variable, and β denotes the state variable. Specifically, the output of each stage is used as the input of the next stage, and the final performance is equal to the expected value of the sum of the costs of each stage. Since the expected value is often not directly calculable, its estimated value is usually generated by the Monte Carlo method. Using the gradient descent algorithm, the parameters of COCP at each stage are trained. The algorithm flow is as follows:

The input of the algorithm is the initial parameters of COCP and the termination condition of the algorithm. In each step, the algorithm calculates the gradient of the objective function with respect to the parameters. When the 2-norm value of the gradient is greater than the termination condition, the parameters are updated and the gradient is recalculated. Iterate in this way until convergence.

When the performance evaluation index converges, the iteration of COCP parameters is automatically completed.

Furthermore, in real-time operation, the trained COCP parameters are directly brought into the approximate value function, and a single-stage deterministic optimization is performed at each stage according to the actual true value of the uncertainty variable (similar to the in-box operation of each stage in Figure 4) to obtain the scheduling strategy for each stage.

In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims (7)

1. A distributed energy robust automatic scheduling method in a multi-stage real-time auxiliary service market is characterized by comprising the following steps:

formulating a deterministic multi-stage optimization model, and defining an optimization target, variables and limiting conditions;

based on the deterministic multi-stage optimization model, a multi-stage distribution robust optimization model is formulated;

under a specific fuzzy set, converting the multi-stage distributed robust optimization model into a net structure, and constructing a dynamic planning model;

and adjusting parameters in the convex optimization control strategy cost function in the dynamic programming model by using a gradient descent algorithm until the performance evaluation index converges to obtain a scheduling strategy.

2. The method for automatically dispatching distributed energy robustness in a multi-stage real-time auxiliary service market according to claim 1, wherein the calculation formula of the multi-stage distributed robust optimization model is as follows:

wherein t represents the decision of the t-th stage, G is the decision variable of the schedule, ζ is the uncertainty variable, Q is the distribution of the uncertainty variable,

for the solution space of decision variables, +.>

Representing mathematical expectations, F (·) represents a cost function.

3. The distributed robust automatic scheduling method in a multi-stage real-time auxiliary service market according to claim 1, wherein the dynamic programming model is converted from a multi-stage distributed robust optimization model, and is in the form of:

wherein t represents the decision of the t-th stage, V _t (. Cndot.) represents the optimization objective of stage t, G is the decision variable of the schedule, ζ is the uncertainty variable, Q is the distribution of uncertainty variables,

for solution space of decision variables, F _t (. Cndot.) represents the current phase cost function, (. Cndot.)>

Representing a cost function of the future stage, and beta represents a state variable.

4. The method for automatically dispatching the distributed energy robustness in the multi-stage real-time auxiliary service market according to claim 1, wherein the most robust distribution of uncertainty variables in the dynamic programming model is specifically expressed as follows:

u _t 、l _t the upper and lower bounds of uncertainty variables, respectively, delta being the dirac function; if mu is _t As an average value of uncertainty variable, weight alpha _t The value of (a) is (u) _t -μ _t )/(u _t -l _t )。

5. The method for robust automatic scheduling of distributed energy sources in a multi-stage real-time auxiliary service market according to claim 1, wherein the single-stage parameterized convex optimization control strategy is obtained by a dynamic programming model and is expressed as:

the form of the convex optimization control strategy cost function is convex quadratic, and is specifically expressed as:

wherein the parameters are

And->

Collectively referred to as COCP parameter, expressed as θ _t+1 T represents the decision of the t-th stage, G is the decision variable of the schedule, Q is the distribution of uncertainty variables, +.>

For solution space of decision variables, F _t (. Cndot.) represents the current phase cost function, (. Cndot.)>

A cost function representing a future phase, β representing a state variable, +.>

Representing the decision variables.

6. The method for automatically dispatching distributed energy robustness in a multi-stage real-time auxiliary service market according to claim 1, wherein a gradient descent algorithm is utilized to adjust a parameter theta= { theta in a convex optimization control strategy cost function generated by a dynamic programming model _t T=1, …, T until the performance evaluation index converges, and a scheduling policy is obtained, and the ith step of updating the gradient descent algorithm is as follows:

θ ⁱ⁺¹ ←Π _Θ (θ ⁱ -α ⁱ g ⁱ )，

wherein g ⁱ For the total cost (J) to theta ⁱ Gradient of theta is theta ⁱ Is pi _Θ (. Cndot.) projection mapping characterizing feasible regions, alpha ⁱ Is the learning rate.

7. The method for automatically dispatching the distributed energy robustness in the multi-stage real-time auxiliary service market according to claim 1, further comprising the step of directly carrying out single-stage deterministic optimization according to the true value of the uncertainty variable in each stage by utilizing the trained COCP parameters to be directly carried into a dynamic programming model, so as to obtain a dispatching strategy of each stage.

Images