CN-122026340-A - AC/DC power system operation reliability assessment method and system based on deep neural network

CN122026340ACN 122026340 ACN122026340 ACN 122026340ACN-122026340-A

Abstract

The invention discloses an AC/DC power system operation reliability assessment method and system based on a deep neural network, which relate to the technical field of power system operation optimization and artificial intelligence and comprise the steps of constructing a multi-state reliability model of an AC/DC power system according to hierarchical progressive logic based on the composition characteristics of the AC/DC power system; the method comprises the steps of sampling system states by adopting a state enumeration method, calculating optimal power flows under different system states, obtaining corresponding optimal cut load amounts, constructing a training data set based on an input feature matrix, constructing an alternating current-direct current power system optimal cut load regression model based on a depth residual error network, training the training data set, carrying out optimal cut load prediction on a state to be evaluated of a target alternating current-direct current power system by utilizing the trained depth residual error network, and calculating a system operation reliability index by combining a prediction result. The method disclosed by the invention ensures the evaluation accuracy and improves the evaluation efficiency at the same time, and provides reliable decision support for the operation scheduling of the AC/DC power system.

Inventors

XIONG KANG
DING YI
BAO MINGLEI
NI YUWEN

Assignees

浙江大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The method for evaluating the operation reliability of the AC/DC power system based on the deep neural network is characterized by comprising the following steps of: Based on the composition characteristics of the AC/DC power system, constructing a multi-state reliability model (100) of the AC/DC power system according to hierarchical progressive logic; sampling system states by adopting a state enumeration method, calculating optimal power flows under different system states, acquiring corresponding optimal cut load amounts, and constructing a training data set based on an input feature matrix; Constructing an optimal cut load regression model (200) of an alternating current-direct current power system based on a depth residual error network, and training a training data set; Performing optimal load shedding prediction on the state to be evaluated of the target AC/DC power system by utilizing the trained depth residual error network, and calculating a system operation reliability index by combining a prediction result; And the depth residual error network stacks the module input and the convolution layer output through jump connection of the residual error module.
2. The method for evaluating the operational reliability of an AC/DC power system based on a deep neural network of claim 1, wherein said multi-state reliability model (100) of the AC/DC power system comprises, Firstly, a direct current equipment reliability model (M1) is built, secondly, a reliability model (M2) of a flexible multi-terminal direct current transmission system is built based on the direct current transmission equipment reliability model, finally, a multi-state reliability model (100) of a wind power plant at a transmitting end and an alternating current power grid at a receiving end is synthesized, and an alternating current and direct current system multi-state reliability model (100) based on flexible transmission is built.
3. The method for evaluating the operational reliability of the deep neural network-based AC/DC power system of claim 2, wherein the modeling (M1) the reliability of the DC power device comprises, The state space is combined based on the available power transmission capacity of the direct current power transmission equipment, and the direct current power transmission equipment is modeled into a three-state model, wherein the three-state model comprises a normal working state (101) with the available power transmission capacity of 100%, a partial failure state (102) with the available power transmission capacity of 50% and a complete failure state (103) with the available power transmission capacity of 0.
4. The method for evaluating the operational reliability of the deep neural network-based alternating current/direct current power system according to any one of claims 1 to 3, wherein the calculating the optimal power flow under different system states and obtaining the corresponding optimal cut load amount comprises, The system states are enumerated through a state enumeration method, an alternating current-direct current system optimal power flow model is solved for each state, the system load shedding cost in the minimized state is taken as an objective function, and the calculation constraint condition of the optimal power flow is considered.
5. The method for evaluating the operational reliability of the deep neural network-based AC/DC power system of claim 4, wherein the constraint condition for calculating the optimal power flow comprises, The method comprises the following steps of direct current system power balance constraint, alternating current system public coupling node power balance constraint, alternating current system power balance constraint, generator set power constraint, node load reduction constraint, node voltage constraint, VSC converter station transmission capacity constraint and alternating current and direct current power constraint.
6. The method for evaluating the operation reliability of the deep neural network-based alternating current-direct current power system according to any one of claims 1 to 3 and 5, wherein the input feature matrix comprises, Node load level, wind farm available capacity, conventional unit available capacity, node self admittance for characterizing line operation status, and dc transmission line available capacity.
7. The method for evaluating the operation reliability of the AC/DC power system based on the deep neural network of claim 6, wherein the deep residual error network comprises, And the residual error modules adopt a jump connection structure, a short circuit path is arranged in parallel by a bypass on the basis of stacking two convolution layers, and a residual error mapping function, namely, the residual error between the input characteristic and the expected output, is learned.
8. The method for evaluating the operational reliability of the deep neural network-based AC/DC power system according to any one of claims 1 to 3, 5 and 7, wherein the optimal cut load regression model (200) comprises, Based on the architecture of the depth residual error network, the training adopts the mean square error as a loss function, and the tangential load quantity is fitted through the mapping relation between the learning system state and the optimal tangential load quantity.
9. The method for evaluating the operational reliability of an AC/DC power system based on a deep neural network of claim 8, wherein said system operational reliability index comprises, The expected lack of power and probability of load shedding; The desired shortage amount of power is expressed as: , Wherein, the In order to average the amount of power load loss, As a total number of system states, For the total number of nodes in the system, Is in a state of Lower part(s) Time node Is a cut load amount; the probability of load loss is expressed as: , Wherein, the For the probability of loss of power load of the system, Is in a state of Lower part(s) Total cut load of the system at the moment.
10. An AC/DC power system operation reliability evaluation system based on a deep neural network adopts the AC/DC power system operation reliability evaluation method based on the deep neural network as claimed in any one of claims 1-9, and is characterized by comprising a reliability modeling module, a data set construction module, a model training module and an evaluation prediction module; The reliability modeling module is used for establishing a multi-state reliability model (100) of the direct current transmission equipment, the flexible multi-terminal direct current transmission system, the transmitting-end wind power plant and the receiving-end alternating current power grid according to the hierarchical progressive logic; The data set construction module is used for enumerating the system states by adopting a state enumeration method, calculating the optimal power flow of the alternating current-direct current system under each state to obtain the optimal cut load quantity, and constructing an input characteristic matrix based on the node load level, the available capacity of a wind power plant, the available capacity of a conventional unit, the node self admittance and the available capacity of a direct current transmission line to form a training data set; The model training module is used for constructing an optimal cut load regression model (200) based on a depth residual error network, learning residual error mapping by utilizing the residual error module and the jump connection structure, and training the model by using a training data set to fit the mapping relation between the system state and the optimal cut load; The evaluation prediction module is used for predicting the optimal cut load quantity of the state to be evaluated of the target AC/DC power system by using the trained depth residual error network model, and calculating the operation reliability index according to the prediction result.

Description

AC/DC power system operation reliability assessment method and system based on deep neural network Technical Field The invention relates to the technical field of operation optimization and artificial intelligence of power systems, in particular to an AC/DC power system operation reliability assessment method and system based on a deep neural network. Background Along with the large-scale grid connection of new energy, the AC/DC hybrid power system plays an important role in promoting the transregional allocation of the new energy. However, the deep coupling of the dc power transmission system and the ac power system leads to an increase in the scale of system elements and the complexity of topology, and the number of expected failure scenarios increases dramatically. The existing reliability evaluation method of the AC/DC power system, whether an analytic method or an analog method represented by Monte Carlo simulation, is seriously dependent on repeated calculation of massive fault scenes. This results in extremely low evaluation efficiency, and it is difficult to meet the requirements for evaluation timeliness in the power system operation phase. In recent years, a data driving method is used for improving the evaluation efficiency, but a traditional machine learning algorithm model represented by a support vector machine is simple in structure and limited in feature learning capability, and evaluation accuracy is difficult to guarantee when processing high-dimensional and strong-nonlinearity power system data. The conventional deep learning method represented by convolutional neural network is easy to cause gradient disappearance and network degradation problems when the number of network layers is increased, so that abnormal phenomena of deeper network and lower precision are caused, and the learning capability of the method on complex fault modes is limited. Therefore, a new evaluation method is needed to greatly improve the evaluation efficiency on the premise of ensuring high precision so as to meet the online decision requirement of the operation stage. Disclosure of Invention The present invention has been made in view of the above-described problems. Therefore, the method solves the technical problems that the existing reliability evaluation method for the AC/DC power system has lower evaluation efficiency, simple algorithm model structure, limited learning ability and how to improve evaluation accuracy. The technical scheme includes that the alternating current-direct current power system operation reliability assessment method based on the deep neural network comprises the steps of constructing a multi-state reliability model of an alternating current-direct current power system according to hierarchical progressive logic based on the characteristics of the alternating current-direct current power system, sampling system states by adopting a state enumeration method, calculating optimal power flows under different system states and obtaining corresponding optimal cut load amounts, constructing a training data set based on an input feature matrix, constructing an alternating current-direct current power system optimal cut load regression model based on a deep residual network, training the training data set, utilizing the trained deep residual network to conduct optimal cut load prediction on the state to be assessed of the target alternating current-direct current power system, and calculating a system operation reliability index according to a prediction result, wherein the deep residual network overlaps module input and convolution layer output through jump connection of a residual module. The multi-state reliability model of the alternating current/direct current power system comprises the steps of firstly establishing a direct current equipment reliability model, secondly establishing a reliability model of a flexible multi-terminal direct current power transmission system based on the direct current power transmission equipment reliability model, and finally integrating the multi-state reliability models of a transmitting-end wind power plant and a receiving-end alternating current power grid to establish the multi-state reliability model of the alternating current/direct current system based on flexible power transmission. The method for evaluating the operation reliability of the alternating current-direct current power system based on the deep neural network is characterized by comprising the steps of establishing a reliability model of direct current equipment, carrying out state combination on a state space based on available power transmission capacity of the direct current power transmission equipment, and modeling the direct current power transmission equipment into a three-state model, wherein the three-state model comprises a normal working state with the available power transmission capacity of 100%, a partial failure state with the available power transmission capacity of 50%,