CN-122000883-A - Method and device for quantifying node voltage uncertainty of source end data driving

CN122000883ACN 122000883 ACN122000883 ACN 122000883ACN-122000883-A

Abstract

The invention discloses a method and a device for quantifying node voltage uncertainty driven by source data, which aim at solving the problems that the phenomenon of right deletion of data commonly existing in a low-voltage station area is ignored and the deviation degree of voltage prediction caused by the complex co-linearity of the source data and user data is large. The method comprises the steps of firstly pushing a source-load aggregation linear sensitivity model based on a physical power flow equation, secondly introducing an inverse probability deletion weighting mechanism to improve a folding knife method model average ridge regression algorithm, realizing that voltage-power sensitivity is identified from biased cut-off data under the condition of no topological parameter, simultaneously, utilizing a source end competition mechanism to jointly identify the phase of a node, and finally, establishing a linear mapping relation from source end power uncertainty to node voltage uncertainty to realize node voltage uncertainty quantification and effectively solve the problems of low-voltage distribution network perception and risk assessment under the condition of limited data and containing photovoltaic.

Inventors

SHI ZHENGYUAN
WANG KAI
LI JUNYI
XIE CHENG
TONG LI
ZENG PINGLIANG
LIN MOHAN
WEI TONG

Assignees

国网浙江省电力有限公司电力科学研究院

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. The source end data driven node voltage uncertainty quantification method is characterized by comprising the following steps of S1, constructing a source end-to-node double-factor linear observation model based on load and photovoltaic variation based on disFlow tide equation: in the formula, And Representing the squares of the node i and source terminal voltages respectively, And The correction sensitivity of the node i to net load fluctuation and photovoltaic feedback characteristics is provided; In the event of a noise error, Is the source end payload fluctuation; Is a typical photovoltaic output; S2, acquiring and preprocessing source end and user side data of a low-voltage station area, and acquiring a preliminary user phase based on a source end competition mechanism; S3, constructing an inverse probability deletion weighting (IPCW) matrix to correct data deviation, combining a Kaplan-Meier-based method to construct weights, forming and applying a weight diagonal matrix, performing sensitivity estimation by adopting a folding-cutter model average ridge regression (JMAR) algorithm, and outputting a minimum statistical branching phase; S4, decomposing the node voltage reconstruction error into background noise and photovoltaic noise, and quantizing the voltage based on a given confidence range through engineering approximate voltage. Preferably, the data in step S2 are preprocessed by the Z-score method standardized.
2. The method for quantifying node voltage uncertainty of source-side data driving according to claim 1, wherein the data in step S2 is preprocessed by a standardized manner of a Z-score method.
3. The method for quantifying node voltage uncertainty of source-side data driving of claim 1, wherein the folding-knife model average ridge regression (JMAR) algorithm of step S3 comprises setting a plurality of And establishing candidate models, adopting a reserved cross-validation evaluation to obtain the average prediction error weight of each model, and obtaining a final predicted value according to the weighted average of the predicted results.
4. A source-side data-driven node voltage uncertainty quantization method according to claim 3, wherein a projection matrix is calculated based on a candidate model, a corresponding vector of prediction error is calculated and represented in a matrix form, an error matrix is constructed, and the corresponding weight is obtained as a quadratic programming problem to solve.
5. The method for quantifying node voltage uncertainty of a source data drive of claim 1 wherein step S2 further comprises constructing a source phase competition model based on data using independent fluctuation characteristics of source three-phase voltages, and initially determining a user phase by minimizing a signal reconstruction residual.
6. The method according to claim 5, wherein step S3 further comprises calculating minimum cross-validation risk of the user belonging to three power lines based on the sensitivity to confirm the phase of the user.
7. The method of claim 1, wherein the step S5 of quantifying the voltage comprises creating a linear mapping from uncertainty of source power to uncertainty of node voltage to achieve uncertainty quantification of node voltage.
8. The method of claim 1, wherein the node voltage uncertainty quantization is performed by a root mean square error.
9. The method for quantifying the uncertainty of the source-side data driven node voltage of claim 8, the method is characterized in that the engineering approximate voltage is in Gaussian distribution.
10. An apparatus for quantifying source-side data-driven low-voltage zone node voltage uncertainty comprising a processor and a memory, characterized in that the processor is configured to read instructions from the memory to perform the method of any of claims 1 to 9.

Description

Method and device for quantifying node voltage uncertainty of source end data driving Technical Field The invention relates to the technical field of operation and control of a power distribution network of a power system, in particular to a method and a device for quantifying node voltage uncertainty driven by source end data. Background The permeability of new energy represented by distributed photovoltaic in a low-voltage distribution network is continuously improved. Up to now, distributed photovoltaics have become an important model of power production and clean energy consumption. However, the distributed photovoltaic output is affected by meteorological factors such as solar irradiance, temperature and the like, and has remarkable intermittence and fluctuation. When high proportion photovoltaic is connected into a low-voltage station area, the safety problems of tidal current foldback, node voltage out-of-limit and the like are extremely easy to be caused. Therefore, the uncertainty of the node voltage is accurately quantified, and the method has important significance for improving the perceptibility and the operation toughness of the power distribution network. At present, methods for analyzing and predicting the voltage of a power distribution network are mainly divided into two types, wherein the first type is a method based on a physical model. The method relies on accurate power distribution network topological structure and line impedance parameters, and solves voltage distribution through tide calculation. However, the problems of file deletion, topology update lag, unknown line parameters and the like commonly exist in the actual low-voltage area, so that the traditional physical model method is difficult to directly apply. The second category is data-driven based methods. In recent years, deep learning algorithms such as long-short-term memory networks (LSTM) and Convolutional Neural Networks (CNN) have been widely used for voltage prediction. But such methods typically fall into a "black box" model, have poor physical interpretability, and require extremely high data quality. Although the prior research has made a certain progress in voltage prediction, in the application scenario of the actual low-voltage area, there is still a key technical problem which is not fully solved. First, the phenomenon of "right-hand deletion" of data is ignored, and most of the existing data driving methods assume that the historical data is complete and real. However, in actual operation, when the voltage of the photovoltaic grid-connected point exceeds the safety threshold, the photovoltaic inverter automatically triggers overvoltage control (such as Volt-Watt control) to perform derating or switching. This results in the collected historical metrology data in which the true value of the high voltage portion is truncated or hidden (statistically referred to as "right-hand-delete"). Training directly with these biased cut-off data can lead to a serious underestimation of the risk of voltage violations by the model, which is of little relevance in the prior art. And secondly, the source end and the user side data have the same linearity, the power fluctuation of the source end (the distribution transformer low-voltage side) is highly coupled with the user side load and the photovoltaic output in a low-voltage area, and the multiple source data have obvious common linearity. In the conventional least square regression method, the parameter estimation variance is extremely large under the condition, and a stable sensitivity coefficient is difficult to obtain. Third, the low voltage area generally has the problem of user phase recording errors or deletions. Most of the existing voltage analysis methods assume that the topology of the area is known, and if topology information is wrong, the calculation of voltage-power sensitivity is directly invalid. In view of the foregoing, there is a need for a method that can accurately quantify voltage uncertainty in the case where topology parameters are unknown and there is a right-hand erasure of measured data due to inverter control. Disclosure of Invention The invention overcomes the defect of larger voltage prediction deviation degree caused by the fact that the phenomenon of right deletion of data is ignored and the complex linearity of the source end and the user side data in the existing platform region assessment mode, and provides a node voltage uncertainty quantification method and device driven by the source end data, which can improve the phenomenon that the complex linearity of the source end and the user side data and the phenomenon of right deletion are ignored, so that the voltage is predicted more accurately, and the requirements on topology information and data fidelity are lower. In order to solve the technical problems, the invention adopts the following technical scheme: a method for quantifying node voltage uncertainty of source end data driving comp