CN114298495A - Power grid reliability evaluation method based on fine-grained data driving - Google Patents

Abstract

The invention belongs to the field of power systems, and particularly relates to a power grid reliability evaluation method based on fine-grained data driving; the method comprises the steps of randomly extracting power grid system state data by using a Monte Carlo method; calculating a power difference value of system direct current power flow before and after the power grid system is switched on and switched off; inputting the power difference value and other power data into a power flow module as input characteristic vectors of a sample to be tested; calculating a regression result and a classification result of the sample to be detected, and determining a problem sample; inputting the input characteristic vector of the problem sample into an optimal power flow module, calculating a classification result of the problem sample, and determining a fault sample; and carrying out minimum load shedding analysis processing on the fault sample, calculating the uncertainty of the fault sample by combining the variance coefficient according to the analysis result, and outputting the reliability evaluation result of the fault sample. The method can be widely applied to the online calculation of the operation reliability of the large power grid, and is particularly suitable for the scene after the large-scale access of high-proportion new energy.

Description

Power grid reliability evaluation method based on fine-grained data driving

Technical Field

The invention belongs to the field of power systems, aims to realize efficient calculation of large power grid reliability evaluation by utilizing a deep neural network, and particularly relates to a power grid reliability evaluation method based on fine-grained data driving.

Background

For high-proportion renewable energy power systems with strong uncertainty, large power grid reliability assessment is an important tool for preventing major power failure accidents. The reliability evaluation process comprises power flow calculation and optimal power flow calculation. In order to obtain high-precision evaluation results, large-grid reliability evaluation needs to analyze a large number of system states (namely, repeatedly executing a large number of power flow and optimal power flow calculation). Therefore, large grid reliability assessment faces the difficult problem of heavy computational burden. The above-mentioned difficulties are further exacerbated by the complexity of the reliability assessment model and its proliferation of computational demand frequency with the access of high-proportion renewable energy sources.

In recent years, with the continuous development of information technology, deep learning technology and the like, data-driven methods have great potential in reducing the computational burden of reliability evaluation, and the methods are increasingly favored by students. Current data-driven methods for reliability assessment typically establish a classifier that reduces the frequency of optimal power flow calculations by identifying whether a sample is marked as "successful" or faulty. After classification, optimal power flow calculation is performed only on samples marked as failures with the goal of minimizing load shedding, and samples marked as "success" do not need to be analyzed. In the reliability evaluation, since most system states belong to the "success" category, the above method can save a lot of computation cost.

However, the existing method is difficult to accurately obtain a high-precision reliability evaluation result, and the main reasons are as follows: the non-linearity of the power flow equation, the discreteness of branch cut-offs, and the imbalance between success/failure samples lead to a complex learning task for reliability evaluation. In the existing method, in order to reduce the learning complexity and realize the reliability evaluation of data driving, the uncertainty of imbalance problems and branch random faults is ignored, or/and the power grid trend is assumed to be linear, so that the reliability evaluation is subjected to great precision loss, and the reliability evaluation is difficult to popularize and apply in the actual industry. Therefore, a more effective data driving method is urgently needed to ensure the accuracy of the reliability evaluation result of the large power grid.

Disclosure of Invention

Based on the problems in the prior art, the invention aims to provide a power grid reliability evaluation method based on fine-grained data driving, aiming at the defects of a data-driven power system operation reliability calculation method. The technical scheme for realizing the invention comprises the following steps:

acquiring state data of a power grid system, and randomly extracting a part of the state data of the power grid system according to the variance coefficient by using a Monte Carlo method;

inputting the power difference value, the load active vector and the reactive vector of a sample to be tested, the maximum available active power and the reactive power of the unit, and the minimum active output and the reactive output of the unit into a power flow module based on data driving;

the power flow regression module calculates a regression result of the sample to be tested, the power flow classification module predicts a classification result of the sample to be tested, and the regression result and the classification result are output as output feature vectors of the sample to be tested;

when the regression result exceeds a corresponding preset threshold value or/and the mark of the classification result is a problem mark, taking the sample to be detected as a problem sample;

inputting the input characteristic vector of the problem sample into an optimal power flow module based on data driving, predicting the classification result of the problem sample, outputting the classification result as the output characteristic vector of the problem sample, and determining the fault sample;

performing minimum load shedding analysis processing on the fault sample, and calculating to obtain the active power of an inflow branch circuit and an outflow branch circuit, and the voltage amplitude and the voltage phase angle;

and according to the calculation result of the minimum load shedding analysis of the fault sample, calculating by combining the variance coefficient to obtain the uncertainty of the fault sample, and outputting the reliability evaluation result of the fault sample.

After the technical scheme is adopted, the invention mainly has the following effects:

1. compared with the existing data-driven operation reliability evaluation method, the data-driven power flow module introduced by the invention can be used as a buffer to effectively identify the problem sample, namely the power flow out-of-limit system state, and reduce the learning complexity of identifying the fault sample. In addition, a part of non-problem samples are identified in advance through the buffer to serve as successful samples, and the problem of sample imbalance can be effectively relieved.

2. The data-driven fine-grained reliability evaluation framework provided by the invention comprises a power flow module and a minimum load shedding calculation model, so that the framework can calculate any reliability index. Meanwhile, through the characteristic vector and the learning strategy designed by the invention, the frame can effectively deal with the random fluctuation of the load/new energy to predict the random fault of the power grid equipment without assuming the linear trend or the unchanged power grid structure, and the calculation precision is high.

3. The method can be widely applied to the online calculation of the operation reliability of a large power grid, and is particularly suitable for the condition after the large-scale access of high-proportion new energy such as wind, electricity and the like.

Drawings

FIG. 1 is a fine-grained operational reliability assessment framework diagram based on a deep neural network in an embodiment of the present invention;

fig. 2 is a flowchart of a power grid reliability evaluation method based on fine-grained data driving in the embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Fig. 1 is a framework diagram for evaluating reliability of a power grid, which is based on a deep learning network, and as shown in fig. 1, the framework diagram mainly includes four parts, firstly, a part of system states are obtained in a sampling manner, and are input into a problem sample identification model driven by data, and in the identification model, a problem sample needs to be found out by using a power flow module driven by data, including a power flow classification module and a power flow regression module driven by data, and then a fault sample is found out from the problem sample. Then, only minimal load shedding analysis is needed to be carried out on the fault sample, and reliability index statistics is carried out according to the analysis result, wherein the sample to be tested is divided into a success sample, a fault sample and a common sample; the normal samples are screened out by a data-driven power flow module, and the remaining problem samples are determined, wherein the problem samples comprise success samples and failure samples.

In contrast to the traditional data-driven reliability assessment framework, the framework of the present invention introduces a buffer (i.e., a data-driven power flow module) to identify problem samples, i.e., system states including line overload, voltage violation, or system disconnection, rather than directly identifying fault samples. The buffer identifies all system states into two stages, wherein the first stage is to distinguish whether the problem samples are problem samples, and the second stage is to distinguish whether the problem samples are fault samples, and the two stages can effectively reduce the learning complexity and relieve the problem of sample imbalance. Considering that the accuracy of the reliability evaluation index of the large power grid may be reduced due to the error identification of the problem sample, the framework combines the data-driven trend regression and the classification result, so that the identification result of the problem sample is more accurate. In particular, if a sample is identified as a problem sample by a data-driven power flow regression model or a data-driven power flow classification model, the sample is regarded as a problem sample.

It can be understood that the above framework alleviates the over-fitting problem to some extent by establishing a buffer to identify the problem sample first, and combines the trend regression result and the trend classification result to further improve the identification accuracy of the problem sample.

In addition, in view of how to construct a nonlinear and discrete feature vector for effectively representing operational reliability, and designing a deep learning strategy adapted to each data-driven module in the framework is a key point for implementing the framework function, fig. 2 shows a flow chart of a power grid reliability evaluation method based on fine-grained data driving in an embodiment of the present invention, and as shown in fig. 2, the evaluation method includes:

101. acquiring state data of a power grid system, and randomly extracting a part of the state data of the power grid system according to the variance coefficient by using a Monte Carlo method;

in the embodiment of the invention, because the state data of the power grid system is very massive and huge, the direct processing of the data is practical, and based on the fact, a Monte Carlo method is adopted to randomly sample a proper amount of system states, and part of system states can be extracted according to the variance coefficient serving as an extraction proportion.

In the embodiment of the present invention, the variance coefficient is expressed as:

wherein α represents a variance coefficient;

an estimated value representing a reliability index;

representing an estimated value

Variance of (2)Here, the reliability index may include a Loss of Load Probability (Loss), an Expected insufficient power (EENS), and the like, and the variance coefficient may be calculated according to an estimated value of the corresponding reliability index, and the corresponding extraction ratio may be guided according to the variance coefficient. The coefficient of variance is used to represent the estimated value

May also be used as one of the convergence criteria of the monte carlo method.

102. Preprocessing the extracted state data of the power grid system, and calculating to obtain a power difference value of the direct current flow of the system before and after the power grid system is switched on and switched off;

in the embodiment of the invention, the extracted power grid system state sample needs to be preprocessed, the preprocessing mode adopts a z-score method as shown in the following formula, and the sample is divided into a training set, a verification set and a test set (to-be-tested sample);

wherein v is_meanAnd v_stdThe mean and standard deviation, respectively, of the vector V.

In the embodiment of the invention, on one hand, the accuracy of sample classification is considered to be the important influence of how to effectively characterize the random fault of the branch circuit. The traditional technology generally adopts available transmission capacity as a characteristic vector for representing the uncertainty of the power grid branch, but the method has difficulty in effectively reflecting the connection relationship between the branches and the importance of the disconnection of a certain branch. In contrast, the invention provides that the power difference value calculated by the direct current power flow of the system before and after the disconnection is used as the characteristic vector for representing the uncertainty of the branch circuit.

Firstly, the invention calculates the power of each branch DC power flow according to the admittance matrix and the voltage phase angle, which is expressed as P_k＝B_kδ_k；

Wherein, P_kPower representing dc power flow on branch k，B_kRepresenting the admittance matrix, δ, on branch k_kDenotes the voltage phase angle difference over the branch k, where k ∈ [1,2]And L represents the total number of branches in the power grid system.

Secondly, the invention calculates the power difference of the direct current flow of different branches before and after the disconnection by transferring the distribution factor DF, which is expressed as

Wherein, DF_k-lRepresenting the change in DC power flow, P, over branch k caused by interruption of branch l_lIndicating the DC power flow on the branch L, L ≠ k, k, L ∈ [1,2]。

Finally, the invention carries out matrix splicing on the power difference of the direct current flow of each different branch before and after the branch is cut off to obtain the power difference vector of the direct current flow of all the branches before and after the branch is cut off, which is expressed as delta P_DC＝P_a-P_b；

Wherein, Δ P_DCPower difference vector, P, representing the DC power flow of all the branches before and after the branch is disconnected_aAnd P_bAnd the direct current power flow vectors of all the branches before and after the branch is disconnected are respectively represented.

It can be understood that, under a certain system state, the change of the power flow of each branch can reflect the change of the admittance matrix B, and further reflect the topological connection relationship of the power grid to a certain extent. In addition, the relative importance of the disconnection of different branches is reflected by the change of the power value of the direct current power flow.

It is worth noting that the method does not need to store the direct current power flow power change values of the branches before and after the different branches are disconnected, which is very helpful for practical application and can reduce calculation power and storage space. The characteristic vectors before and after the different branches are disconnected can be updated on line by transferring the distribution factor DF, the processing of the global state data of the power grid system can be avoided, the global burden can be effectively reduced, the calculation cost and the storage cost are smaller than those of global operation, and the method is better suitable for large-scale power grid reliability evaluation.

103. Acquiring a power difference value, a load active vector and a reactive vector of a sample to be detected, the maximum available active power and reactive power of a unit, and the minimum active power output and reactive power output of the unit, and inputting the power difference value, the load active vector and the reactive vector, the maximum available active power and the reactive power of the unit, and the minimum active power output and the reactive power output of the unit into a power flow module based on data driving as input characteristic vectors of the sample to be detected;

after the system state data is preprocessed, a power flow equation and a minimum tangential load optimization model can be respectively solved based on a Newton method and an interior point method to solve a load active vector and a reactive vector, a maximum generatable active power and a reactive power of a unit, a minimum active power and a reactive power of the unit and a power difference value of a system direct current power flow before and after the power grid system is disconnected.

The power flow equation is expressed as:

wherein, P_i(V, δ) represents the injected active power at node i, Q_i(V, δ) represents the injected reactive power at node i; v_iIs the voltage amplitude at node i, V_jIs the voltage amplitude at node j, and V represents the voltage amplitude vector; n represents the total node number of the power grid system; g_ijIs the real part of node i to node j in the node admittance matrix, B_ijIs the imaginary part from node i to node j in the node admittance matrix; delta_ijRepresenting the voltage phase angle difference, δ, from node i to node j_ij＝δ_i-δ_j，δ_iRepresenting the phase angle, δ, of the voltage at node i_jThe voltage phase angle at node j is represented and δ represents the voltage phase angle matrix.

The minimum load shedding model is expressed as:

s.t.

P_i(V,δ)-PD_i+C_i＝0,i∈ND

Q_i(V,δ)-QD_i＝0,i∈ND

0≤C_i≤PD_i,i∈ND

V_i ^min≤V_i≤V_i ^max,(i∈N)

wherein, C_iRepresenting the amount of off-load at node i; ND represents the number of load nodes of the grid system, PD_iAnd QD_iRespectively representing active load and reactive load at a node i; PG (Picture experts group)_iAnd QG_iThe active power and the reactive power of the node i are generated respectively;

is the minimum active output of the unit at node i;

is the maximum active output of the unit at node i;

is the minimum reactive power output of the unit at node i,

is the maximum reactive power output of the unit at node i, NG represents the number of all generator nodes in the system, T_k(V, delta) represents the transmission power on branch k,

representing the maximum transmission power on branch k, L representing the number of all branches of the grid system, V_i ^minIs the minimum voltage amplitude, V, at node i_i ^maxIs the maximum voltage magnitude at node i.

In the embodiment of the invention, after a part of state data of a power grid system is extracted, each part of data can be used as a sample to be tested, because an input characteristic vector of a deep neural network needs to comprise information such as load, new energy, topological structure and the like of a large power grid, the input information is relatively complex, in order to be processed simply and conveniently, a power flow equation and a minimum load shedding optimization model are respectively solved based on a Newton method and an interior point method, a power difference value, a load active vector and a reactive vector, maximum generatable active power and reactive power of a unit and minimum active power and reactive power of the unit before and after the power grid system is disconnected are obtained and are used as input, and an input characteristic vector is formed and is input into a power flow module based on data driving, wherein the input characteristic vector X of reliability evaluation is expressed as:

X＝[P_D,Q_D,PG^max,QG^max,PG^min,QG^min,ΔP_DC]

wherein, the input feature vector X is the input feature vector P of the sample to be tested of the present invention_DAnd Q_DRespectively a load active vector and a load reactive vector; PG (Picture experts group)^maxAnd QG^maxThe maximum active power and the maximum reactive power of the generator set are respectively generated; PG (Picture experts group)^minAnd QG^minThe lowest active output and the lowest reactive output of the generator set are respectively.

Note that the new energy is regarded as a load in the present invention. The system load and available generation capacity are configured as input eigenvectors to characterize uncertainty of new energy, load demand, and generator state.

It can be understood that, in the embodiment of the present invention, on the premise of inputting the feature vector X, the power flow regression equation is used to solve in combination with the input data of the input feature vector X, and the power flow regression module is used to obtain the voltage amplitude vector of the bus node, the branch power vector, the output of the unit on the balance node, that is, the regression result of the active power vector and the reactive power vector output by the unit.

104. The power flow regression module calculates a regression result of the sample to be tested, the power flow classification module predicts a classification result of the sample to be tested, and the regression result and the classification result are output as output feature vectors of the sample to be tested;

in the embodiment of the invention, the data-driven power flow module comprises a power flow regression module and a power flow classification module, the data-driven power flow module is a deep learning network, and the power flow regression module and the power flow classification module can also be sub-modules of the deep learning network and are mainly used for calculating a regression result and predicting a classification result.

In the embodiment of the invention, the regression result calculated by the power flow regression module is used for representing the node voltage of all buses, the power of all lines and the output power of the generator, and the classification result predicted by the power flow classification module is used for distinguishing whether the sample to be tested is a problem sample.

In the embodiment of the invention, the construction of the output feature vector is mainly determined by each data driving module. Thus, the output feature vector of the reliability assessment is represented as:

Y_PF＝[V,T,PG_ref,Q_G,F_PF]

wherein, Y_PFRepresenting an output feature vector of a sample to be tested; v is a bus node voltage amplitude vector; t is the branch power vector; PG (Picture experts group)_refIs the active power vector of the unit output on the balance node; q_GIs a reactive power vector output by the unit; f_PFThe flag vector, which is the problem sample, takes either 0 or 1.

It will be appreciated that the vectors V, T, PG_ref,Q_GIs output by the power flow regression module, and F_PFAnd if the power flow classification module outputs the power flow classification module, the power flow classification module and the power flow classification module form an output characteristic vector of the power flow module based on data driving.

105. When the regression result exceeds a corresponding preset threshold value or/and the mark of the classification result is a problem mark, taking the sample to be detected as a problem sample;

in the embodiment of the invention, the accuracy of the reliability evaluation index of the large power grid is reduced by considering that the error identification of the problem sample may cause the reduction of the accuracy of the reliability evaluation index of the large power grid, so that the identification result of the problem sample is more accurate by combining the data-driven power flow regression result and the power flow classification result. Specifically, if a certain sample to be tested is identified as a problem sample by the power flow regression module or the power flow classification module, the sample to be tested is regarded as the problem sample, for example, if the regression result of the voltage amplitude is 1.8p.u., and the preset threshold of the voltage amplitude is 1.05p.u., then obviously, the regression result of the voltage amplitude already exceeds the corresponding preset threshold, and then the sample to be tested at this time will be used as the problem sample to perform subsequent secondary classification.

In the embodiment of the present invention, the preset threshold is a preset engineering value, and a person skilled in the art can set the corresponding threshold according to actual conditions, which is not specifically limited by the present invention.

106. Inputting the input characteristic vector of the problem sample into an optimal power flow module based on data driving, predicting the classification result of the problem sample, outputting the classification result as the output characteristic vector of the problem sample, and determining the fault sample;

because only the problem sample is screened in the process, but whether the problem sample is a fault sample is unknown, the problem is further classified and identified by adopting the optimal power flow module based on data driving, so that the problem can be further solved by taking the corresponding input feature vector as the input feature vector in the optimal power flow module based on data driving based on the classification result of the problem sample, and the method can be expressed as follows:

X₁＝[P_D,Q_D,PG^max,QG^max,PG^min,QG^min,ΔP_DC]

wherein the content of the first and second substances,

here, the input feature vector X₁I.e. the input feature vector of the problem sample of the present invention.

In the embodiment of the invention, in order to quickly identify the fault sample, the optimal power flow module based on data driving is not provided with the power flow module, only the classification module is arranged, and in order to ensure that the classification module can effectively identify the fault sample from the problem sample, the output characteristic vector is set as Y_OPF＝[F_OPF]Wherein F is_OPFAnd taking 0 or 1 for the flag vector of whether the system state is a fault system.

It can be understood that, in the embodiment of the present invention, on the premise of inputting the feature vector X of the problem sample, the minimum load shedding model is used to solve in combination with the input data of the input feature vector X, and the optimal power flow module driven by data determines whether the current node is load shedding, and obtains the result of the flag vector of the fault system state.

Because the data-driven power flow module and the data-driven optimal power flow module are both formed by the deep neural network, the deep neural network is explained in a unified way, and one of the most frequently used activation functions in the deep learning field, namely the ReLU activation function, is adopted in the input layer and the middle layer of the deep neural network. For the activation function of the last layer of neural network, a linear activation function is selected, which enables the neural network to capture a wider output. Therefore, for the data driving module for reliability evaluation, the mapping relationship between the input and the output can be expressed by the following formula.

Wherein, Y_PFRepresenting an output feature vector of a sample to be tested;

representing a weight matrix of the sample to be tested between the i-th layer neural network and the i + 1-th layer neural network;

represents the activation function of the tested sample X in the i-th layer neural network,

representing a bias vector of the sample to be detected between the i-th layer neural network and the i + 1-th layer neural network; 1,2, n, n represents the number of neural network layers; theta_PF＝{W_PF,b_PF}；Y_OPFAn output feature vector representing a problem sample;

representing a weight matrix of the problem sample between the ith layer neural network and the (i + 1) th layer neural network;

represents the activation function of the problem sample X at the i-th layer neural network,

representing the bias vector of the problem sample between the i-th layer neural network and the i + 1-th layer neural network; 1,2, n, n represents the number of neural network layers; theta_OPF＝{W_OPF,b_OPF}。

In order to obtain the optimal deep neural network parameter theta, namely theta_PFAnd theta_OPFThe mean square error function is typically used as the loss function L, expressed as:

wherein N represents the number of samples;

and y_iAre the actual and predicted values for the ith sample.

Then the corresponding optimization objective is to minimize L. The above unconstrained optimization problem can be solved correspondingly by using an Rmsprop or Adam algorithm.

In the preferred embodiment of the invention, a load shedding sample sensitive loss function is provided in an intelligent classification module aiming at a success sample/fault sample, namely a success/fault system state, so as to improve the learning precision of a deep neural network on a fault sample classified in error. Specifically, the proposed load shedding sensitive loss function is as follows:

L_OPF＝βL₁+L₂

wherein L is₁A loss function representing successful samples, i.e. the mean square error of the successful sample classification result; l is₂A loss function representing a fault sample, namely the mean square error of a fault sample classification result; omega_sRepresents a successful sample set, Ω_fA set of fault samples is represented. The parameter β is a user-defined weight, ranging from (0, 1). By this loss function, L can be minimized_OPFTo ensure the overall accuracy of classification; meanwhile, the system state is more sensitive to the error prediction fault by adjusting the weight beta.

It should also be understood that, in the embodiment of the present invention, a power flow equation and a minimum tangential load optimization model may be respectively solved based on a newton method and an interior point method, so as to obtain a power difference value, a load active vector and a reactive vector, a maximum generatable active power and a reactive power of a unit, and a power difference value between a minimum active output and a reactive output of the unit and a direct current power flow of a system before and after a power grid system is disconnected as an input feature vector; meanwhile, branch power, voltage amplitude, unit output, a power flow out-of-limit mark, a load shedding mark and the like are reserved and used as output characteristic vectors, the input characteristic vectors are input into a power flow module, a problem sample is obtained through prediction by combining the result of the output characteristic vectors, the input characteristic vectors are input into an optimal power flow module, and a fault sample is obtained through prediction by combining the result of the output characteristic vectors.

107. Performing minimum load shedding analysis processing on the fault sample, and calculating to obtain the active power of an inflow branch circuit and an outflow branch circuit, and the voltage amplitude and the voltage phase angle;

in the embodiment of the present invention, since a fault sample is determined, minimum load shedding analysis processing needs to be performed on the fault sample, and the calculating to obtain the active power of the incoming branch and the outgoing branch, and the voltage amplitude and the voltage phase angle includes:

s.t.

P_i(V,δ)-PD_i+C_i＝0,i∈ND

Q_i(V,δ)-QD_i＝0,i∈ND

0≤C_i≤PD_i,i∈ND

V_i ^min≤V_i≤V_i ^max,(i∈N)

wherein, C_iRepresenting the amount of off-load at node i; ND represents the number of load nodes of the grid system, P_i(V, δ) represents the injected active power at node i, Q_i(V, δ) represents the injected reactive power V at node i_iIs the voltage amplitude at node i, δ_iRepresenting the phase angle, δ, of the voltage at node i_jRepresenting the phase angle of the voltage at node j, δ_ijRepresenting the voltage phase angle difference, δ, from node i to node j_ij＝δ_i-δ_j；PD_iAnd QD_iRespectively representing active load and reactive load at a node i; PG (Picture experts group)_iAnd QG_iThe active power and the reactive power of the node i are generated respectively; g_ijIs the real part of node i to node j in the node admittance matrix, B_ijIs the imaginary part from node i to node j in the node admittance matrix;

is the minimum active output of the unit at node i;

is the maximum active output of the unit at node i;

is the minimum reactive power output of the unit at node i,

is the maximum reactive power output of the unit at node i, NG represents the number of all generator nodes in the system, T_k(V, delta) represents the transmission power on branch k,

representing the maximum transmission power on branch k, L representing the number of all branches of the grid system, V_i ^minIs the minimum voltage amplitude, V, at node i_i ^maxIs the maximum voltage amplitude at node i, and N represents the total number of nodes in the system.

108. And according to the calculation result of the minimum load shedding analysis of the fault sample, calculating by combining the variance coefficient to obtain the uncertainty of the fault sample, and outputting the reliability evaluation result of the fault sample.

In the embodiment of the present invention, the reliability index statistics may be performed according to the calculation result of the minimum load shedding analysis of the fault sample, and the reliability index mainly considered in the present invention includes: a Load of Load Proavailability (LOLP), and a desired power shortage (EENS). Meanwhile, the method can be used as the convergence condition of the Monte Carlo method according to both the variance coefficient alpha and the number of sampling samples.

In order to better illustrate the evaluation scheme of the invention, the evaluation is performed in combination with specific power grid system state data, and in the embodiment, the validity of the invention is verified by adopting an IEEE RTS-79 system, an IEEE RTS-96 system and modified examples thereof. The example information is as follows:

example 1: an IEEE RTS-79 system. Assuming that the load curve follows a normal distribution with a standard deviation of 10% of the expected value, the historical annual peak load for the IEEE RTS-79 system is expected.

Example 2: EEE RTS-96 System. In this system, three IEEE RTS 79 systems are connected by six transmission lines.

Example 3: a modified IEEE RTS-79 system. The permeability of renewable energy is 20%.

Example 4: a modified IEEE RTS-96 system. The permeability of renewable energy is 20%.

Convergence of Monte Carlo methodProvided that when the variance coefficient α is less than 5% or the number of sampling samples reaches 100000. The Monte Carlo method is combined with the Newton-Raphson method and the interior point method to solve the reliability result which is an accurate solution. The hyper-parameters of the deep neural network are detailed in table 1. The following indexes are used to quantify the accuracy of the load flow calculation. P_vmIt is a ratio that the absolute error of the voltage amplitude exceeds 0.001 p.u. P_vaIt is the ratio of the absolute error of the voltage phase angle exceeding 0.01 rad. P_pf/P_qtThe absolute error of the active power of the inflow/outflow branch is over 5 MW. M _ VM, M _ VA, M _ PF, and M _ PT correspond to the root mean square error of the above variables.

TABLE 1 hyper-parameters of neural network intelligent modules of different depths under different arithmetic examples

The accuracy of a classifier is typically measured by the following criteria: ACC, SEN, SPE, g-mean.

Wherein

Wherein

Table 2 shows the data of different examplesAnd (5) driving a power flow regression result. It can be seen from Table 2 that the root mean square error M _ VM of the voltage amplitude does not exceed 5.00X 10 in any of examples 1 to 4^-4p.u.; the root mean square error M _ VA of the voltage phase angle is not more than 5.00 multiplied by 10 in any of the embodiments 1 to 4^-3rad; the root mean square error of the branch power includes that neither M _ PF nor M _ PT exceeds 2.22 MW. In addition, the probability indexes of voltage and power are less than 1.5% in each of the embodiments 1 to 4. Therefore, the calculation accuracy of all the concerned variables is high due to the strong approximation capability of the deep neural network.

TABLE 2 data-driven regression results of power flow under different calculation examples

Based on the high-precision power flow regression result, the problem sample can be identified by comparing the high-precision power flow regression result with the operation limit value of the power system. Table 3 lists the results of classifying the problem samples by different methods under different examples. As shown in table 3, the overall Accuracy (ACC) of the data-driven trend regression and classification model for problem sample classification was over 99.45% and 98.85%, respectively, under all the examples. Notably, SEN quantifies the proportion of correctly classified problem samples to the total number of problem samples. The larger the SEN value, the fewer problem samples that are misclassified. Using data-driven regression or classification results alone, the minimum values of SEN in examples 1-4 were 99.45% and 99.17%, respectively. By combining the regression result and the classification result driven by data, namely taking the union set of the problem samples finally judged by the regression model and the classification model driven by data, the minimum value of the SEN can be improved to 99.43% and 99.82% while high precision (ACC) is kept, so that a more reliable classification result is obtained, and the effectiveness of the method is verified.

TABLE 3 results of classifying problem samples by different methods under different examples

Table 4 lists the classification results for success/failure samples in different cases. As can be seen from the number of problem samples and the number of failure samples in the table, the problem of sample imbalance is very serious under all the calculation examples. In addition, it can also be intuitively seen that the imbalance ratio of success samples to failure samples can be reduced by first identifying the problem samples rather than directly classifying all samples as being successful/failed. When the conventional loss function is directly used, the ACC of all the examples can reach more than 95%, but the SEN values are 97.91%, 94.05%, 98.68% and 93.65%, respectively. And the influence of each fault sample on the reliability index is not negligible. In this block, the larger the value of SEN, the fewer the number of erroneous judgment failure samples. Clearly, the effectiveness of the success/failure sample identification is not satisfactory using the conventional loss function, especially for examples 2 and 4, with values below 95%. When the loss function provided by the invention is adopted, the SEN values of the calculation examples 1-4 can be respectively increased to 99.42%, 97.99%, 99.02% and 95.92% as can be seen from the table 4, and the SEN values of each calculation example are improved to be more than 95%. Since the imbalance ratio of success/failure samples is different under different calculations, the setting of the weight β is also different. The threshold selected in this chapter is 0.4 when sorting. The results show that the load shedding sensitive loss function can effectively improve the identification precision of the fault sample to more than 95%, and meets the requirement of industrial application precision.

TABLE 4 results of classification of "success/failure" samples under different examples

To verify the advancement of the present invention, the performance comparison results are shown in table 5, compared to the most advanced data-driven reliability evaluation method. As can be seen from table 5, the relative error between lopp and EENS calculated by MLKNN is much larger than the minimum relative error of MLKNN calculated by the proposed method, which is much larger than the maximum relative error of the proposed method. The main reason is because the proposed method has divided the problem into sub-tasks to significantly reduce the complexity. In addition, a specific learning strategy is further proposed to ensure the accuracy of the calculation. Furthermore, the proposed method shows faster computational performance due to the direct mapping function of the deep neural network. MLKNN, however, must compare a large number of samples to each other, which is time consuming. In summary, the proposed method represents a significant advantage compared to the state of the art data driven methods.

Compared with the reference value, from the viewpoint of calculation accuracy, the reliability index estimated by the method is very close to an accurate value as can be seen from 0 through high-precision identification of the fault sample by the deep neural network. The relative errors of LOLP and EENS are no more than 3.3% and 0.45%, respectively. From a computation time point of view, the proposed method greatly reduces the computation time by avoiding unnecessary power flow optimization. The acceleration ratio of the proposed method ranges from 6.2 to 23.4. Therefore, when transmission line faults and complex alternating current power flows are considered at the same time, the data driving framework provided by the chapter can quickly and accurately calculate the reliability index of the large power grid.

TABLE 5 Performance comparison of the present invention with the classical method of reliability assessment of large grids

In summary, the invention provides a power grid reliability evaluation method based on fine-grained data driving, and compared with the existing data-driven operation reliability evaluation method, the introduced buffer can effectively identify the problem sample, namely the system state with the tidal current exceeding the limit, and the learning complexity for identifying the fault sample is reduced. A part of successful samples are identified in advance through the buffer, and the problem of sample imbalance can be effectively relieved. In addition, the reliability evaluation framework provided by the invention can improve the calculation speed by 4-23 times under the condition of meeting the calculation precision requirement. Compared with the conventional data driving method, the method can comprehensively consider the alternating current load flow, the random faults of the branch circuits and the like, and the calculation accuracy is further guaranteed.

In the description of the present invention, it is to be understood that the terms "coaxial", "bottom", "one end", "top", "middle", "other end", "upper", "one side", "top", "inner", "outer", "front", "center", "both ends", and the like, indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of description and simplicity of description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, are not to be construed as limiting the present invention.

In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "disposed," "connected," "fixed," "rotated," and the like are to be construed broadly, e.g., as meaning fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; the terms may be directly connected or indirectly connected through an intermediate, and may be communication between two elements or interaction relationship between two elements, unless otherwise specifically limited, and the specific meaning of the terms in the present invention will be understood by those skilled in the art according to specific situations.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (5)

1. A power grid reliability assessment method based on fine-grained data driving is characterized by comprising the following steps:

acquiring state data of a power grid system, and randomly extracting a part of the state data of the power grid system according to the variance coefficient by using a Monte Carlo method;

preprocessing the extracted state data of the power grid system, and calculating to obtain a power difference value of the direct current flow of the system before and after the power grid system is switched on and switched off;

acquiring a power difference value, a load active vector and a reactive vector of a sample to be detected, the maximum available active power and reactive power of a unit, and the minimum active power output and reactive power output of the unit, and inputting the power difference value, the load active vector and the reactive vector, the maximum available active power and the reactive power of the unit, and the minimum active power output and the reactive power output of the unit into a power flow module based on data driving as input characteristic vectors of the sample to be detected;

the power flow regression module calculates a regression result of the sample to be tested, the power flow classification module predicts a classification result of the sample to be tested, and the regression result and the classification result are output as output feature vectors of the sample to be tested;

when the regression result exceeds a corresponding preset threshold value or/and the mark of the classification result is a problem mark, taking the sample to be detected as a problem sample;

inputting the input characteristic vector of the problem sample into an optimal power flow module based on data driving, predicting the classification result of the problem sample, outputting the classification result as the output characteristic vector of the problem sample, and determining the fault sample;

performing minimum load shedding analysis processing on the fault sample, and calculating to obtain the active power of an inflow branch circuit and an outflow branch circuit, and the voltage amplitude and the voltage phase angle;

and according to the calculation result of the minimum load shedding analysis of the fault sample, calculating by combining the variance coefficient to obtain the uncertainty of the fault sample, and outputting the reliability evaluation result of the fault sample.

2. The fine-grained data-driven power grid reliability assessment method according to claim 1, wherein the variance coefficient is expressed as:

wherein α represents a variance coefficient;

an estimated value representing a reliability index;

representing an estimated value

The variance of (c).

3. The fine-grained data-driven power grid reliability evaluation method according to claim 1, wherein the calculating of the power difference of the system direct current flows before and after the power grid system is disconnected comprises calculating the power of each branch direct current flow, represented as P, according to the admittance matrix and the voltage phase angle_k＝B_kδ_k(ii) a Calculating the power difference of the direct current flow of different branches before and after the disconnection by transferring the distribution factors, and expressing the power difference as

And obtaining power difference vectors of the direct current power flows of all the branches before and after the branch is cut off, wherein the power difference vectors are expressed as delta P_DC＝P_a-P_b；

Wherein, P_kRepresenting the power of the DC current flow on branch k, B_kRepresenting the admittance matrix, δ, on branch k_kRepresenting the phase angle of the voltage on branch k,

representing the change in power of the DC current on branch k caused by interruption of branch l, DF_k-lRepresenting the change in DC power flow, P, over branch k caused by interruption of branch l_lIndicating the DC power flow on the branch L, L ≠ k, k, L ∈ [1,2]L represents the total number of branches in the grid system, Δ P_DCPower difference vector, P, representing the DC power flow of all the branches before and after the branch is disconnected_aPower vector, P, representing the DC power flow of all branches before the branch is disconnected_bAnd the power vector represents the direct current power flow of all the branches after the branches are disconnected.

4. The fine-grained data-driven-based power grid reliability evaluation method according to claim 1, wherein the data-driven-based optimal power flow module is trained by using a load shedding sensitive loss function, wherein the load shedding sensitive loss function is expressed as:

L_OPF＝βL₁+L₂

wherein L is_OPFRepresenting the loss function of the shear load sensitivity, beta representing a weighting factor, L₁A loss function representing a successful sample is generated,

N₁indicates the number of successful samples, Ω_sA successful set of samples is represented as a set of samples,

actual value, y, representing the ith problem sample_iA predictor representing an ith problem sample; l is₂A loss function representing a sample of the fault,

N₂represents the number of fault samples, Ω_fA set of fault samples is represented.

5. The method according to claim 1, wherein the performing minimum load shedding analysis on the fault sample and calculating the active power of the incoming branch and the outgoing branch, and the voltage amplitude and the voltage phase angle comprises:

s.t.

P_i(V,δ)-PD_i+C_i＝0,i∈ND

Q_i(V,δ)-QD_i＝0,i∈ND

0≤C_i≤PD_i,i∈ND

wherein, C_iRepresenting the amount of off-load at node i; ND represents the number of load nodes of the grid system, P_i(V, δ) represents the injected active power at node i, Q_i(V, δ) represents the injected reactive power at node i; v_iIs the voltage amplitude at node i, δ_iRepresenting the phase angle, δ, of the voltage at node i_jRepresenting the phase angle of the voltage at node j, δ_ijRepresenting the voltage phase angle difference, δ, from node i to node j_ij＝δ_i-δ_j；PD_iAnd QD_iRespectively representing active load and reactive load at a node i; PG (Picture experts group)_iAnd QG_iThe active power and the reactive power of the node i are generated respectively; g_ijIs the real part of node i to node j in the node admittance matrix, B_ijIs the imaginary part from node i to node j in the node admittance matrix;

is the minimum active output of the unit at node i;

is the maximum active output of the unit at node i;

is the minimum reactive power output of the unit at node i,

is the maximum reactive power output of the unit at node i, NG represents the number of all generator nodes in the system, T_k(V, delta) represents the transmission power on branch k,

representing the maximum transmission power on branch k, L representing the number of all branches of the grid system, V_i ^minIs the minimum voltage amplitude, V, at node i_i ^maxIs the maximum voltage amplitude at node i, and N represents the total number of nodes in the system.

Images