CN-122001010-A - Method for constructing robust scheduling model of water-wind-solar complementary system

CN122001010ACN 122001010 ACN122001010 ACN 122001010ACN-122001010-A

Abstract

The invention provides a method for constructing a robust scheduling model of a water-wind-solar complementary system, which is used for constructing a multi-source heterogeneous data set by fusing meteorological data, equipment state data, power grid load data and historical operation data, generating an artificial intelligent model based on the data set training, generating a climate-energy-load-equipment state composite extreme scene conforming to a physical rule, constructing an uncertainty set containing the composite extreme scene, establishing a two-stage robust optimization framework embedded with physical constraints, and adopting a column and constraint generation algorithm to carry out iterative solution on the robust optimization framework to generate a scheduling scheme considering economy and safety. The method can break through the dependence of the traditional method on historical data, generate a multi-dimensional compound extreme scene conforming to the physical rule, remarkably improve the dispatching robustness and reliability of the water-wind-solar complementary system under extreme working conditions, and realize the dynamic balance of economy and safety.

Inventors

CAO JINGWEI
YANG CHENGBI
ZHAO SIYI
ZHANG DEBAO
LONG JIAN
HU FENG
QIU SHI
MA YUEJIAO
DENG LIWEI
YANG LONGBAO
LIU XIN
CHEN CHUAN
LI QING

Assignees

中国华能集团清洁能源技术研究院有限公司
华能澜沧江水电股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260108

Claims (10)

1. The method for constructing the robust scheduling model of the water-wind-solar complementary system is characterized by comprising the following steps of: s1, acquiring meteorological data, energy operation data, load data and equipment state data, and constructing a training data set containing a climate-energy-load-equipment state coupling relation based on cleaning alignment and characteristic engineering of multi-source heterogeneous data; S2, training a generated AI model based on the training data set, wherein the generated AI model is input by taking a historical meteorological mode, equipment operation working conditions and a load level as conditions, and a multidimensional compound extreme operation scene with physical rationality is generated through countermeasure type or denoising type training; S3, constructing the multi-dimensional composite extreme operation scene into an extreme scene uncertainty set, and establishing a two-stage robust optimization framework considering multi-dimensional uncertainty by combining physical operation constraint of a water-wind-light system, an equipment aging model and a power grid safety criterion; And S4, carrying out iterative solution on the two-stage robust optimization framework by adopting a column and constraint generation algorithm, generating a basic scheduling decision and a worst-case adjustment strategy through interaction of a main problem and a sub-problem, and finally outputting a pre-decision scheduling scheme which takes economy and robustness into consideration.
2. The method of claim 1, wherein S1 comprises: S11, carrying out space-time alignment processing on the meteorological data, mapping discrete meteorological observation point data to a unified space-time grid through an interpolation algorithm, and eliminating time sequence dislocation caused by data acquisition time interval difference; S12, carrying out health status quantification processing on the equipment status data, and calculating the equipment health attenuation rate by adopting the equipment aging index.
3. The method of claim 1, wherein S2 comprises: s21, adopting a condition generation countermeasure network architecture, taking a historical meteorological mode, equipment operation working conditions and a load level as condition input, and generating an extreme scene meeting physical rationality constraint through countermeasure training of a discriminator and a generator; S22, generating an extreme scene through a denoising process of the diffusion model, and gradually converting the initial noise vector into a scene conforming to physical constraints.
4. The method of claim 1, wherein S3 further comprises: s31, embedding an equipment aging model into an uncertainty set for construction, and dynamically adjusting equipment output boundary constraint through an aging index; s32, introducing a power grid safety criterion as a scene screening condition, and performing feasibility verification on the generated extreme scenes to ensure that all scenes meet a power grid safety constraint set.
5. The method of claim 1, wherein S4 comprises: S41, adopting an improved column and constraint generation algorithm, generating a basic scheduling decision through iteration of a main problem, and searching a worst scene based on a sub-problem; S42, setting a convergence acceleration mechanism, extracting a representative scene subset through scene clustering in each iteration, and reducing the scale of the sub-problem to reduce the computational complexity.
6. The utility model provides a water scene complementary system robust scheduling model construction device which characterized in that includes: the data acquisition and processing module acquires meteorological data, energy operation data, load data and equipment state data, and builds a training data set containing a climate-energy-load-equipment state coupling relation based on cleaning alignment and characteristic engineering of multi-source heterogeneous data; The generation type AI training module is used for training a generation type AI model based on the training data set, the generation type AI model is input by taking a historical meteorological mode, equipment operation working conditions and a load level as conditions, and a multidimensional composite extreme operation scene with physical rationality is generated through countermeasure type or denoising type training; The set construction and optimization framework module is used for constructing the multi-dimensional composite extreme operation scene into an extreme scene uncertainty set, and establishing a two-stage robust optimization framework considering multi-dimensional uncertainty by combining physical operation constraint of a water-wind-light system, an equipment aging model and a power grid safety criterion; And the solving and strategy generating module adopts a column and constraint generating algorithm to carry out iterative solving on the two-stage robust optimization framework, generates a basic scheduling decision and a worst-case adjusting strategy through interaction of the main problem and the sub-problem, and finally outputs a pre-decision scheduling scheme which takes economy and robustness into consideration.
7. The apparatus of claim 6, wherein the data acquisition and processing module is further to: carrying out space-time alignment processing on the meteorological data, mapping discrete meteorological observation point data to a unified space-time grid through an interpolation algorithm, and eliminating time sequence dislocation caused by the difference of data acquisition time intervals; And carrying out health status quantification processing on the equipment status data, and calculating the equipment health attenuation rate by adopting the equipment aging index.
8. The apparatus of claim 6, wherein the generated AI training module is further to: The method comprises the steps of adopting a condition generation countermeasure network architecture, taking a historical meteorological mode, equipment operation working conditions and a load level as condition input, and generating an extreme scene meeting physical rationality constraint through countermeasure training of a discriminator and a generator; generating an extreme scene through a denoising process of the diffusion model, and gradually converting an initial noise vector into a scene conforming to physical constraints.
9. A computer device comprising a processor and a memory; Wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, for implementing a method for constructing a robust scheduling model of a water-wind-solar complementary system according to any one of claims 1 to 5.
10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements a method of building a robust scheduling model for a hybrid wind and solar system according to any of claims 1-5.

Description

Method for constructing robust scheduling model of water-wind-solar complementary system Technical Field The invention belongs to the technical field of water-wind-light complementary systems, and particularly relates to a method for constructing a robust scheduling model of a water-wind-light complementary system. Background Currently, along with the continuous improvement of the renewable energy source duty ratio, the optimization scheduling research of the water-wind-solar centralized control center is changed from the traditional single energy source independent optimization to the full-system multi-energy cooperative complementation, and the scheduling complexity is exponentially increased. However, the existing scheduling technology has obvious theoretical and method bottlenecks in coping with increasingly frequent extreme working conditions, and severely restricts the toughness and reliability of system operation. The core architecture of the existing mainstream scheduling model is still severely dependent on historical observation data and conventional probability distribution (such as normal distribution and Weber distribution) to predict new energy output and load demand, whether the existing mainstream scheduling model is based on a deterministic optimization algorithm or a random planning method considering uncertainty, and the data driving paradigm is applicable under normal conditions of stable climate environment and stable system operation, but the inherent defects are not revealed under extreme working conditions. First, these models consider a serious deficiency for the small probability, high destructive extreme events of the "black swan" and "gray rhinoceros" formulas. Such events include, but are not limited to, extreme weather events (e.g., very large floods in centuries impacting hydropower stations, continuous windless weather resulting in total wind power outages, severe cold or hot waves causing severe load climbs), sudden failures of equipment triggered by recessive defects or aging accumulation (e.g., main transformer explosions, turbine blade breaks), and critical control system failures caused by network attacks, among others. Existing models either fail to learn effectively due to the scarcity of such samples in the historical data, or ignore the correlation and linkage effects between these events due to the assumption that the conditions are too ideal. As a result, the optimal scheduling strategy, which is excellent in normal state, may fail rapidly once the above-mentioned extreme conditions are met, so that not only the optimal goal cannot be achieved, but also the accident may be amplified due to the wrong scheduling instruction (such as excessive dependence on water and electricity in the water shortage period), and systematic breakdown may be caused. Secondly, although methods such as robust optimization have been developed in the academia to deal with uncertainty, the existing robust optimization method still has a great deal of distraction when dealing with uncertainty of multi-dimension and strong coupling of a water-wind-solar complementary system. The fundamental limitation is the coarseness and stiffness of the uncertainty set construction method. The current method mostly relies on simple interval estimation (such as 'wind and light output is +/-30% of a predicted value') or extrapolation based on limited historical extreme samples, and has two major defects that firstly, it is difficult to accurately describe the dynamic coupling relation between uncertainty of three dimensions of climate-energy source and load, for example, extreme drought weather can simultaneously lead to photovoltaic output enhancement (due to sunlight enhancement), water and electricity output sharp reduction (due to water reduction) and load change (due to irrigation requirement), and the complex relevance cannot be embodied in a simple interval model, and secondly, the method lacks 'imagination' of novel extreme working conditions which are not available but are physically possible, for example, certain novel compound extreme events (such as strong typhoon is accompanied by continuous overcast, wind power is simultaneously limited by extreme cutting and overlooking) in the climate change background, and the scope of historical data is exceeded. The prior art mainly comprises two types, namely a random planning model based on a scene analysis method and a robust optimization model adopting a traditional uncertainty set. Scene analysis approximately describes uncertainty by generating a large number of discrete scenes, but the generation of a scene library is severely limited by historical data or simple Monte Carlo simulation, and extreme scenes with physical reality exceeding historical experience cannot be creatively generated, so that the scene library is insufficient in protection against unknown risks. While conventional robust optimization models, while providing a "worst-case" feasi