CN-121984140-A - Value function deduction correction method applied to rolling unit combination problem

CN121984140ACN 121984140 ACN121984140 ACN 121984140ACN-121984140-A

Abstract

The invention discloses a value function deduction correction method applied to a rolling unit combination problem, which comprises the following steps of firstly establishing a rolling unit combination model based on a Markov decision process, secondly solving the rolling unit combination model under a frame similar to dynamic programming, carrying out deduction on the decision by using a Bayesian inference mechanism to realize dynamic correction of the value function, improving the decision accuracy, and thirdly directly obtaining an optimal solution by using an analytic expression method to obviously improve the calculation efficiency. The invention can dynamically adjust the value function by utilizing the real-time operation information, and improves the scheduling adaptability of the power system under the conditions of uncertainty and randomness, thereby enhancing the economy and the safety of the system, meeting the operation requirement of the new energy under the background of large-scale access, and having obvious application value and social and economic benefits.

Inventors

ZHU JIANQUAN
SONG RUI
WANG SHENGJIE
DING YUJIE
HUANG HAOJIANG
FU GUOBIN
LUO YUHAO
CHEN JIAJUN
HU YUFAN
LI KAI
Lu zhaolong
WANG XUEBIN

Assignees

华南理工大学
国网青海省电力公司电力科学研究院
国网青海省电力公司

Dates

Publication Date: 20260505
Application Date: 20251229

Claims (10)

1. The value function deduction correction method applied to the rolling unit combination problem is characterized by comprising the following steps of: s1, establishing a rolling unit combination model based on a Markov decision process; s2, solving a rolling unit combination model by applying an approximate dynamic programming algorithm, and deducting a decision by utilizing a Bayesian inference mechanism to realize dynamic correction of a value function; S3, directly obtaining the optimal solution by using an analytic expression method.
2. The method of claim 1, wherein the rolling mill set combination model based on the markov decision process includes state variables, decision variables, state transition equations, external random information, and objective functions.
3. The value function deduction correction method applied to the rolling unit combination problem according to claim 2 is characterized in that the constraint of the Markov decision process comprises that the start-stop state of a generator set needs to meet the constraint of minimum start-up time and minimum shutdown time, the active and reactive output of the generator set needs to meet the constraint of corresponding upper and lower limits, the output change of the generator set needs to meet the constraint of climbing rate in adjacent scheduling periods, the supply and demand balance constraint of active power and reactive power needs to be met in each scheduling period by a system, the voltage amplitude of each node needs to be kept in an allowable range, the power transmission of a power transmission line needs to meet the limit of line capacity, and the up-and down-regulation reserve capacity needs to be reserved in the whole system.
4. The method for deducing and correcting the value function applied to the rolling unit combination problem according to claim 2, wherein the state variables comprise unit start-stop time in the current period, unit start-stop state and generating capacity in the previous period, and the decision variables comprise unit start-stop state, active output, reactive output, energy storage charge and discharge capacity, node voltage and phase angle.
5. The method for deducting and correcting the value function applied to the rolling unit combination problem according to claim 2, wherein the state transfer equation describes a state transfer process of the unit between different periods, and the objective function is to minimize the running cost of the unit in the whole period.
6. The method for performing value function deduction correction on a rolling mill set assembly problem according to claim 2, wherein the external random information comprises random change information between the active load and the predicted value of the active load, random change information between the reactive load and the predicted value of the reactive load, and random change information between the predicted values of the wind power generator and the wind power generator.
7. The method for correcting value function deduction applied to rolling unit combination problem according to claim 1, wherein the step S2 comprises the steps of: modeling a similarity function as probability distribution containing mean and uncertainty in the rolling scheduling process; Evaluating the feasible decisions based on the current system state utilization and generating virtual decisions; deducing a state after the decision corresponding to the virtual decision according to the state transfer relation, and pre-correcting the value function by Bayesian reasoning in combination with actual operation feedback information; And then re-evaluating and determining the real decision of the current scheduling moment based on the pre-corrected value function, and continuously updating the value function after obtaining the actual operation feedback, thereby realizing the online self-adaptive optimization of the rolling unit combination decision.
8. The method for deducing and correcting a value function applied to a rolling unit combination problem according to claim 1, wherein the method for directly obtaining the optimal solution by using the analytical expression method in step S3 comprises the following steps: the decision is decomposed into the power generation power and the start-stop strategy, the optimal start-stop strategy is obtained through state deduction, then the optimal start-stop strategy is substituted into a cost function and constraint conditions to construct an optimization problem taking the power generation power as a variable, the optimal solution of the power generation power is obtained through analysis of a Lagrange function and optimality conditions, and the optimal solution of the power generation power is substituted into the cost function to directly obtain the optimal instant cost and deduction cost, so that efficient and rapid decision is realized.
9. A computer device comprising a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, wherein the computer program, when executed by the processor, causes the processor to implement the method as claimed in any one of claims 1 to 8.
10. A computer readable storage medium storing a computer program, wherein the computer program is executed by a processor, the processor implementing the method according to any one of claims 1 to 8.

Description

Value function deduction correction method applied to rolling unit combination problem Technical Field The invention belongs to the field of power system dispatching, and particularly relates to a value function deduction correction method applied to a rolling unit combination problem. Technical Field The safety restraint unit combination plays an important role in guaranteeing the economical efficiency and the safety of the power system. Conventional crew combinations typically formulate an operational plan based on deterministic information. However, with the rapid development of renewable energy sources, the uncertainty of the output makes it difficult for the traditional day-ahead unit combination scheme to meet the actual operation requirements. Therefore, it is especially necessary to develop a rolling safety constraint unit combination model capable of flexibly coping with uncertainty, and the rolling safety constraint unit combination model is gradually and widely focused. The rolling machine set combination is a typical mixed integer nonlinear programming problem, and the solving method mainly comprises three types of classical optimization methods, model predictive control and approximate dynamic programming. Classical optimization methods, such as branch-and-bound methods and the bendes decomposition method, have certain advantages in handling complex constraints, but as the control period increases, the calculation amount increases exponentially, and it is difficult to meet the real-time requirements of large-scale problems. The model prediction control solves the rolling unit combination problem continuously through rolling prediction information, so that the calculation efficiency is improved, but the performance of the model prediction control is highly dependent on the prediction information, and the effect is obviously reduced when the prediction is inaccurate. The approximate dynamic programming method is based on the Belman equation to decompose the problem into multi-stage sub-problems, and the coupling relation among the stages is described by using a value function, so that the balance between the calculation efficiency and the solution precision can be obtained. However, existing approximate dynamic programming methods rely on value functions that are trained offline through historical or simulated data to make online decisions. When the uncertainty of the actual operating environment is inconsistent with the offline assumption, the reliability of the value function decreases, thereby affecting the decision accuracy (Liang Li. Urban road impedance model based on vehicle trajectory and path recommendation method study [ D ]. Jilin university, 2020.). Disclosure of Invention The invention provides a value function deduction correction method applied to a rolling unit combination problem, which aims to solve two technical problems of how to self-adapt a correction value function through a decision deduction mechanism in an online decision stage so as to improve decision quality and how to accelerate a decision deduction process on the premise of ensuring calculation accuracy. The invention is realized at least by one of the following technical schemes. A value function deduction correction method applied to a rolling unit combination problem comprises the following steps: s1, establishing a rolling unit combination model based on a Markov decision process; s2, solving a rolling unit combination model by applying an approximate dynamic programming algorithm, and deducting a decision by utilizing a Bayesian inference mechanism to realize dynamic correction of a value function; S3, directly obtaining the optimal solution by using an analytic expression method. Further, the rolling mill set combination model based on the Markov decision process comprises state variables, decision variables, state transition equations, external random information and objective functions. Further, constraints of the Markov decision process include that the start-stop state of the generator set needs to meet minimum start-up time and minimum shutdown time constraints, the active and reactive power output of the generator set needs to meet corresponding upper and lower limit constraints, the output change of the generator set in adjacent scheduling periods needs to meet climbing rate constraints, the system needs to meet supply and demand balance constraints of active power and reactive power in each scheduling period, the voltage amplitude of each node needs to be kept within an allowable range, the power transmission of a power transmission line needs to meet line capacity constraints, and the system integrally needs to reserve up-and-down-regulation reserve capacity meeting operation safety requirements. Further, the state variables comprise the unit start-stop time in the current period, the unit start-stop state and the generating capacity in the last period, and the decision variables comprise the unit sta