CN-121981513-A - Main-auxiliary micro multidimensional operation risk sensing method considering uncertainty of multiple time scales

CN121981513ACN 121981513 ACN121981513 ACN 121981513ACN-121981513-A

Abstract

The invention relates to the field of safety early warning of power systems, and discloses a main-auxiliary micro multidimensional operation risk sensing method considering multi-time scale uncertainty, which comprises the steps of solving safety constraint of each uncertainty scene and N-1 fault based on a tide optimization model, constructing a probability-based risk assessment method by taking standby shadow price and minimum cut load as risk quantification indexes, and acquiring a daily risk sensing result by carrying out probability weighting and expected calculation on triggering indexes of various risks; and aiming at the identified daily critical risk scene, adopting a neural network prediction model based on resource distribution, a real-time index evaluation model based on online tide and an N-1 rapid classification model based on critical section identification to rapidly sense various risks. The invention has the advantage of realizing rapid and accurate sensing of the multi-dimensional operation risk of the main micro-domain in the day.

Inventors

GAO HONGJUN
YE MENG
NIU ZHENYONG
WANG RENJUN
GUO MINGHAO
WANG YUNLONG
YANG XINYI
HU BOWEN
SONG JINGYI
XU XIAO
LUO LONGBO

Assignees

四川大学

Dates

Publication Date: 20260505
Application Date: 20251201

Claims (8)

1. The main and auxiliary micro multidimensional operation risk sensing method considering the uncertainty of multiple time scales is characterized by comprising the following steps, S1, aiming at the comprehensive awareness of the daily multi-dimensional operation risk, constructing a power flow optimization model I to perceive the daily power electric quantity unbalance risk in a single scene, constructing a power flow optimization model II to perceive the high-low voltage out-of-limit/heavy overload risk in the single scene, constructing a power flow optimization model III to perceive the N-1 fault risk, taking the standby shadow price and the minimum cut load as indexes of risk quantification, and acquiring a daily risk perception result risk assessment method by carrying out probability weighting and expected calculation on trigger indexes of various risks based on a probability risk assessment method; s2, aiming at daily multi-scene risk perception, training a neural network model, and identifying the worst daily key risk scene on line; s3, aiming at the identified daily key risk scene, a neural network prediction model based on resource distribution, a real-time index evaluation model based on online power flow and an N-1 rapid classification model based on key section identification are adopted to rapidly sense various risks.
2. The method for sensing the main and auxiliary multidimensional operation risk taking account of uncertainty of multiple time scales as recited in claim 1, wherein the objective function of the tide optimization model I is: ; wherein, C is the total running cost of the system, T is the total scheduling period, T is a single period in the scheduling period; 、、 i, j and k are indexes of the conventional generator set, the distributed power supply and the energy storage system respectively; The power generation cost function of the conventional generator set i is as follows; active output of the conventional generator set i in a t period; operating cost or benefit function for distributed power source j; Active power output of the distributed power source j in a t period; an operation cost function of the energy storage system k; 、 respectively charging and discharging power of the energy storage system k in a t period; constraint conditions of the tide optimization model I comprise node power balance constraint, conventional unit/distributed power supply output constraint, energy storage operation constraint, network safety constraint and system standby constraint; the daily power and electricity unbalance risk is perceived through a model resolvable criterion and a standby constraint shadow price criterion; The model resolvable criterion is that if the power flow optimization model has no solution in a given day-ahead uncertainty scene, all adjustable resources in the power system cannot meet electric quantity balance and preset standby requirements at the same time, and the system deterministically has an electric power and electric quantity unbalance risk in the scene; The shadow price criterion of the standby constraint is that if a trend optimization model has a solution in a given day-ahead uncertainty scene, the shadow price of the standby constraint of the system is analyzed, if the shadow price is high, the risk degree of electric power unbalance is high, and if the shadow price is zero or low, the risk degree of electric power unbalance is low.
3. The master join in part multidimensional running risk sensing method taking into account multi-time scale uncertainty as recited in claim 1, wherein sensing high and low voltage out-of-limit/heavy overload risk in the single scenario comprises: Carrying out safety check if the calculated power flow calculation is converged, traversing the voltage of all nodes and the power flow of all key branches in the power grid, judging that the scene is temporarily free of risk if all the voltage and power flow indexes are within preset safety limit values, and judging that the scene is running risk if any one or more indexes are out of limit; Based on risk depth quantification of the optimization model, analyzing a scene judged to have high-level static safety risk or running risk through a power flow optimization model II, wherein an objective function of the power flow optimization model II is as follows: ; In the formula, A set of all load nodes in the power system; The constraint conditions of the power flow optimization model II comprise revised node power balance constraint, network safety constraint and control variable constraint; And if the total cut load quantity is larger than zero, the risk can be defined as a deterministic risk, and the specific value of the minimum cut load quantity is a direct quantitative index for the high-low voltage/heavy overload risk severity in the scene.
4. The master join in part multidimensional running risk awareness method of claim 1 wherein the awareness of N-1 failure risk includes: Constructing an N-1 fault scene library, constructing a comprehensive N-1 fault scene library by combining historical operation data analysis and a device reliability database, and performing risk weighting on N-1 fault scenes and probability information by utilizing a typical fault mode of a fault tree analysis refinement key section and the occurrence probability thereof; Based on N-1 risk perception of a transfer supply path, aiming at a transmission ring network/high-voltage closed-loop network line, adopting a direct analysis method based on load flow redistribution to conduct risk analysis, immediately executing conventional load flow calculation once in an initial topology after a fault, judging whether chain overload or voltage overrun occurs to other lines or equipment due to the fact that the transferred load flow is accepted, if so, judging that the N-1 scene is at risk, aiming at a main transformer, adopting a state evaluation method based on preset bus connection operation to conduct risk analysis, further modifying network topology on the basis of the initial topology after the fault, simulating corresponding bus connection switch to be closed, executing conventional load flow calculation once under the new topology after the bus connection transfer supply, judging whether the residual main transformer receiving the load is overloaded again and whether the relevant bus voltage meets the requirement, if not, judging that the N-1 scene is at risk, aiming at a network distribution main line, adopting a restoration capacity evaluation method based on network reconstruction to conduct risk analysis, determining one or more backup transfer supply paths according to the network topology and the preset switch positions, if the backup power supply paths are not one, selecting one backup path, selecting a backup switch to be opened, carrying out conventional load flow calculation once, judging whether the overload is in the pre-selected topology has a low, and if the fault load restoration area is in the pre-selected, and if the fault load has the fault has occurred, and if the fault has been restored, and if the fault has the problem is low, and if the fault is calculated, and if the fault is low.
5. The master join in part multidimensional running risk perception method taking account of multi-time scale uncertainty as recited in claim 4 wherein the perception of N-1 risk of failure further comprises N-1 risk quantification based on an optimization model; constructing a power flow optimization model III, wherein the objective function of the power flow optimization model III is as follows: ; In the formula, A set of all load nodes; Is a node The active load quantity to be cut off is located; Is the load The constraint conditions of the power flow optimization model III comprise a switch state variable and network topology constraint, a power flow constraint for taking the switch state into account, a node voltage constraint, a generator output constraint and a cut load quantity constraint.
6. The master join in part multidimensional running risk awareness method of claim 1 wherein the constructing of the risk assessment method comprises: construction of uncertainty scene set and probability assignment, and typical operation scene set Each scene in the collection Are given corresponding occurrence probabilities And meet the following Constructing an N-1 fault scene library aiming at N-1 risk Each fault is Is given its probability of occurrence ; Risk category triggering criteria and cost quantization indexes, and risk triggering indexes: ; In the formula, Is a binary index for judging the scene At risk Whether the following is in an unsafe state; probability attribution and explicit judgment of daily multidimensional risk based on scene set Probability of And risk indexes, carrying out probability quantification on each type of definite risk, and calculating various risks in a day-ahead scheduling period In an uncertainty scenario Probability of occurrence next; risk probability of unbalance of electric power and electric quantity : ; In the formula, For scenes Whether the power-down electric quantity unbalance risk is triggered or not; High and low voltage out-of-limit/heavy overload risk probability : ; In the formula, For scenes Whether the lower high-low voltage/heavy overload out-of-limit risk is triggered or not; N-1 risk probability : ; In the formula, Representing scenes And malfunction in Whether the N-1 risk is triggered when the occurrence happens; and calculating expected values of various risk consequences in a day-ahead scheduling period, wherein the expected values comprise the following components: risk expected cost of power-electricity imbalance : ; In the formula, Is a scene An average value of Lagrangian multipliers of system reserve constraint in an optimal solution of a lower tide optimization model I; High and low voltage out-of-limit/heavy overload risk expected cut load amount : ; In the formula, Is a scene The total cut load quantity optimal solution of the lower tide optimization model II; N-1 risk desired cut load : ; In the formula, Is a scene And malfunction And (3) combining the total cut load quantity optimal solution of the power flow optimization model III.
7. The master join in part multidimensional running risk awareness method of claim 1 wherein the step of S2 comprises: S21, a key scene offline label fast generation criterion comprises a power electric quantity unbalance risk key scene identification criterion, a high-low voltage out-of-limit/heavy overload risk key scene identification criterion and a risk identification key scene identification criterion which is not met by N-1; s22, a neural network model overall architecture is used for respectively training a power electric quantity unbalance risk identification model, a high-low voltage out-of-limit/heavy overload risk identification model and an N-1 unsatisfied risk identification model aiming at power electric quantity unbalance risks, high-low voltage out-of-limit/heavy overload risks and N-1 unsatisfied risks; S23, marking the worst operation scene through a discrete data set to perform offline training on the power and electricity unbalance risk identification model, the high-low voltage out-of-limit/heavy overload risk identification model and the N-1 unsatisfied risk identification model, and outputting a key risk scene.
8. The master join in part multidimensional running risk awareness method of claim 1 wherein the step of S3 comprises: S31, taking into account the intelligent sensing of the unbalance risk of the power and the electric quantity in the day of regulating and controlling the resource distribution, constructing a neural network prediction model based on the key risk scene of the unbalance of the power and the electric quantity in the day and the adjustable resource distribution, wherein the input of the neural network prediction model is key risk scene information and the regulating and controlling resource distribution in the system, and the unbalance value of the power/electric quantity of the main and micro system is obtained through the output of the neural network prediction model; S32, adopting a real-time index evaluation model based on online power flow calculation to realize real-time evaluation of risks by calculating an overall voltage out-of-limit risk index, a line heavy overload risk index, a transformer heavy overload risk index and a reverse heavy overload risk index of the system; S33, key section identification mining N-1 risk intelligent sensing adopts an N-1 rapid classification model based on key section identification, wherein the model is a multi-layer sensing machine, the input is the combination of worst operation scene characteristics and specific key section identification, and the output is a binary classification result of whether a system is safe or not after the section has N-1 faults.

Description

Main-auxiliary micro multidimensional operation risk sensing method considering uncertainty of multiple time scales Technical Field The invention relates to the field of safety precaution of electric power systems, in particular to a main-matching micro-multidimensional operation risk sensing method considering multi-time scale uncertainty. Background In recent years, with the rapid increase of novel loads such as distributed power sources (such as photovoltaic and wind power) large-scale access, electric automobiles and the like, and the increasingly complex coupling operation of a main network, a distribution network and a micro network, the operation risk of the power system has the characteristics of multidimensional, strong coupling and high uncertainty. The traditional safety analysis method exposes two core challenges when facing the main-allocation micro-domain operation risk (such as unbalanced electric power and electric quantity, heavy overload, high-low voltage out-of-limit and unsatisfied N-1), namely firstly, in the day-ahead stage, huge uncertainty scenes generated by load and new energy prediction errors need to be comprehensively considered, and the system is subjected to empty high-dimensional evaluation, but the traditional certainty analysis cannot meet the accurate quantification requirements on the risk probability and the consequences, and secondly, in the day-ahead stage, the rapid change of the system operation state brings up extremely high real-time requirements (15-minute rolling and second-level response) on the risk perception, and the calculation time of power flow optimization or N-1 verification based on a complete physical model is too long, so that the rapid decision is difficult to support. The current risk perception technology either simplys the model to result in insufficient conservation or is too complex to calculate to meet real-time. Therefore, at the present stage, a main-matching micro-multidimensional operation risk intelligent sensing method considering multi-time scale uncertainty needs to be designed to solve the problems. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides a main-matching micro multi-dimensional operation risk sensing method considering multi-time scale uncertainty. The object of the invention is achieved by a method for master-slave multidimensional running risk perception taking into account multi-time scale uncertainty, comprising the steps of, S1, aiming at the comprehensive awareness of the daily multi-dimensional operation risk, constructing a power flow optimization model I to perceive the daily power electric quantity unbalance risk in a single scene, constructing a power flow optimization model II to perceive the high-low voltage out-of-limit/heavy overload risk in the single scene, constructing a power flow optimization model III to perceive the N-1 fault risk, taking the standby shadow price and the minimum cut load as indexes of risk quantification, and acquiring a daily risk perception result risk assessment method by carrying out probability weighting and expected calculation on trigger indexes of various risks based on a probability risk assessment method; s2, aiming at daily multi-scene risk perception, training a neural network model, and identifying the worst daily key risk scene on line; s3, aiming at the identified daily key risk scene, a neural network prediction model based on resource distribution, a real-time index evaluation model based on online power flow and an N-1 rapid classification model based on key section identification are adopted to rapidly sense various risks. Specifically, the objective function of the power flow optimization model I is: ; wherein, C is the total running cost of the system, T is the total scheduling period, T is a single period in the scheduling period; 、、 i, j and k are indexes of the conventional generator set, the distributed power supply and the energy storage system respectively; The power generation cost function of the conventional generator set i is as follows; active output of the conventional generator set i in a t period; operating cost or benefit function for distributed power source j; Active power output of the distributed power source j in a t period; an operation cost function of the energy storage system k; 、 respectively charging and discharging power of the energy storage system k in a t period; constraint conditions of the tide optimization model I comprise node power balance constraint, conventional unit/distributed power supply output constraint, energy storage operation constraint, network safety constraint and system standby constraint; the daily power and electricity unbalance risk is perceived through a model resolvable criterion and a standby constraint shadow price criterion; The model resolvable criterion is that if the power flow optimization model has no solution in a given day-ahead uncertainty scene, all