CN-122026514-A - Thermal power generating unit dispatching optimization method based on multiple uncertainties

CN122026514ACN 122026514 ACN122026514 ACN 122026514ACN-122026514-A

Abstract

The invention discloses a thermal power unit dispatching optimization method based on multiple uncertainties, which solves the problems that the conventional thermal power unit dispatching does not consider weather and load uncertainties, and the thermal power unit suffers from lost income, running safety risk or difficult dispatching execution, and comprises the steps of collecting the initial data of the thermal power unit dispatching and preprocessing; the method comprises the steps of constructing a multi-uncertainty-physical constraint two-dimensional fusion feature based on preprocessed original data, compressing the original dimension, normalizing the original dimension into a three-dimensional tensor, establishing a prediction data set, establishing a deep learning model fused with CNN and a transducer, outputting a load and weather prediction result with a confidence interval by adopting a multi-split site prediction mechanism, establishing a multi-objective random optimization model based on the prediction result, and solving to obtain an optimal dispatching scheme of the thermal power generating unit. And the load and weather uncertainty are quantified, so that the safe and efficient operation of the thermal power unit is ensured, and the energy consumption and equipment loss of the thermal power unit are reduced while the income is ensured.

Inventors

LI HANQIU
Chai Zhenqi
WANG YIRAN
Liu Weidao
HOU PENG
JIN RONGSEN
YU YANG
HUANG ZILI

Assignees

浙江浙能数字科技有限公司
浙江省白马湖实验室有限公司

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. A thermal power generating unit scheduling optimization method based on multiple uncertainties is characterized by comprising the following steps: S1, collecting dispatching original data of a thermal power generating unit and preprocessing; S2, constructing a multi-uncertainty-physical constraint two-dimensional fusion feature based on the preprocessed original data, compressing the original dimensions, normalizing the original dimensions into a three-dimensional tensor, and building a prediction data set; S3, establishing a deep learning model fusing CNN and a transducer, and outputting a load and weather prediction result with a confidence interval by adopting a multi-split point prediction mechanism; and S4, based on the prediction result, establishing a multi-objective random optimization model considering multiple uncertainties and solving to obtain an optimal scheduling scheme of the thermal power generating unit.
2. The thermal power generating unit dispatching optimization method based on multiple uncertainties according to claim 1, wherein in the step S4, a combined scene set containing multiple uncertainties is randomly generated according to a prediction result, a multi-objective random optimization model containing physical constraint conditions of the thermal power generating unit is built by using economic benefit maximization and cost minimization as double objective functions, and double objectives are converted into single objective optimization through weighted summation.
3. The thermal power generating unit scheduling optimization method based on multiple uncertainties according to claim 1 or 2, wherein the optimal scheduling scheme comprises a scheduling scheme corresponding to each quotation period, wherein the scheduling scheme comprises a thermal power generating unit running state, a target output curve, a fuel consumption strategy, a thermal stress control target and a dynamic response matching scheme.
4. The thermal power unit dispatching optimization method based on multiple uncertainties according to claim 2, wherein the thermal power unit physical operation constraints comprise total output constraint, climbing constraint, output segmentation constraint, thermal stress constraint, thermal efficiency and energy consumption constraint, minimum stable output constraint, start-stop constraint and dynamic response constraint.
5. The thermal power generating unit dispatching optimization method based on multiple uncertainties according to claim 1, wherein the deep learning model is output in a mode of 0.05 minute, load predicted values corresponding to 0.5 minute and 0.95 minute, weather predicted values and electricity price predicted values, and the three predicted results are provided with 90% confidence intervals.
6. The thermal power generating unit dispatching optimization method based on multiple uncertainties according to claim 1, wherein in the step S3, the CNN branch structure comprises a convolution layer, a pooling layer and a flattening layer, the convolution layer extracts local trend and fluctuation features through sliding of a plurality of convolution kernels, the pooling layer reduces feature dimensions, extracts dominant forms, and the flattening layer expands high-dimensional features into one-dimensional feature vectors.
7. The thermal power generating unit dispatching optimization method based on multiple uncertainties according to claim 1, 5 or 6, wherein in the step S3, the transducer branch structure comprises a multi-head attention layer, a multi-layer random inactivation layer and a residual error connection module, the multi-head attention layer adaptively identifies weights of key moments according to input sequences, and the multi-layer random inactivation layer dynamically discards part of neurons in a training process.
8. The thermal power unit dispatching optimization method based on multiple uncertainties according to claim 1, 2, 4, 5 or 6 is characterized in that the thermal power unit dispatching original data comprise weather data, load data and unit physical characteristic data, the weather data comprise historical weather data and corresponding predicted weather data, influences of the weather uncertainties on new energy output, electricity utilization load and thermal power unit operation efficiency are quantized, the load data comprise historical load data and corresponding predicted data, the load prediction uncertainties are quantized, and a time sequence sample is constructed based on the thermal power unit dispatching original data.
9. The thermal power generating unit scheduling optimization method based on multiple uncertainties according to claim 1 or 2 or 4 or 5 or 6, wherein the output of the multi-objective stochastic optimization model comprises output-price ratio pairs, and a stepped non-decreasing price ratio curve is constructed.
10. The thermal power generating unit dispatching optimization method based on multiple uncertainties according to claim 1, 5 or 6, wherein in the step S3, the characteristics extracted by the CNN branch structure and the characteristics extracted by the Transformer branch structure are input into a full-connection layer for fusion after feature splicing.

Description

Thermal power generating unit dispatching optimization method based on multiple uncertainties Technical Field The invention relates to the technical field of thermal power generating unit operation scheduling, in particular to a thermal power generating unit scheduling optimization method based on multiple uncertainties. Background Thermal power generating unit dispatching plays a core role in a modern power system, and the thermal power generating unit dispatching plays a role far exceeding the traditional stable power supply, and becomes a key support for guaranteeing energy transformation and maintaining the safety and economic operation of a power grid. The scheduling operation of the thermal power generating unit is affected by interleaving of multiple uncertainty factors, and the factors are accurately controlled, so that the core premise of making a reasonable thermal power generating unit scheduling scheme is that weather uncertainty (such as extremely high temperature, heavy rain, high wind and the like) can directly affect new energy power generation output (wind power, photovoltaic) and power utilization load (cooling and heating loads), and load prediction uncertainty is caused by industrial production fluctuation, resident power utilization habit change and the like, so that power demand is difficult to accurately estimate. However, the existing thermal power generating unit dispatching optimization does not consider multiple uncertainty factors and physical operation constraints in a system, and obvious limitations exist. For example, the patent CN115833262A discloses a thermal power unit day-ahead dispatching method and device for coordinating wind power uncertainty by water and electricity, a thermal power unit day-ahead dispatching model considering wind power output uncertainty and water and electricity coordination is constructed, and a day-ahead dispatching plan of the thermal power unit is obtained by solving, so that the advantages of water and electricity can be fully utilized, and the problems of strong fluctuation, large prediction error and the like of wind power generation can be overcome. But does not systematically quantify the uncertainty range of weather and load, does not output the probability distribution of electricity price prediction, does not relate to the natural law related requirements of unit combustion efficiency, load matching, start-stop process thermodynamic constraint and the like, and has insufficient adaptability to multiple uncertainty superposition scenes caused by extreme weather. Disclosure of Invention The invention aims to solve the problems that in the prior art, weather and load uncertainty are not considered in thermal power unit scheduling, so that solving efficiency is low, the thermal power unit faces lost benefits, running safety risks or scheduling execution difficulties, and provides a thermal power unit scheduling optimization method based on multiple uncertainties, which quantifies the load and weather uncertainty, ensures safe and efficient running of the thermal power unit, and reduces energy consumption and equipment loss of the thermal power unit while guaranteeing benefits. In order to achieve the above purpose, the present invention adopts the following technical scheme: a thermal power generating unit dispatching optimization method based on multiple uncertainties comprises the following steps: S1, collecting dispatching original data of a thermal power generating unit and preprocessing; S2, constructing a multi-uncertainty-physical constraint two-dimensional fusion feature based on the preprocessed original data, compressing the original dimensions, normalizing the original dimensions into a three-dimensional tensor, and building a prediction data set; S3, establishing a deep learning model of a double-branch structure fused with CNN and a transducer, and outputting a load and weather prediction result with a confidence interval by adopting a multi-branch point prediction mechanism; and S4, based on the prediction result, establishing a multi-objective random optimization model considering multiple uncertainties and solving to obtain an optimal scheduling scheme of the thermal power generating unit. According to the thermal power generating unit dispatching optimization method based on multiple uncertainties, the CNN-converter deep learning model is adopted to synchronously quantify multiple uncertainties of weather and load, random scene generation and multi-objective optimization are combined, risks can be precisely quantified, unit physical constraints are integrated, and expected profit and operation safety balance can be guaranteed under the multiple uncertainty superposition scene. The nonlinear associated characteristics of multiple uncertainty factors are captured through deep learning, the output confidence interval can directly reflect prediction reliability, the random scene generation is combined to break through sample li