CN115392697A - Data-model hybrid driven power system safety assessment method and system - Google Patents
Data-model hybrid driven power system safety assessment method and system Download PDFInfo
- Publication number
- CN115392697A CN115392697A CN202211018209.1A CN202211018209A CN115392697A CN 115392697 A CN115392697 A CN 115392697A CN 202211018209 A CN202211018209 A CN 202211018209A CN 115392697 A CN115392697 A CN 115392697A
- Authority
- CN
- China
- Prior art keywords
- day
- power
- ahead
- data
- power system
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/04—Circuit arrangements for ac mains or ac distribution networks for connecting networks of the same frequency but supplied from different sources
- H02J3/06—Controlling transfer of power between connected networks; Controlling sharing of load between connected networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention provides a data-model hybrid-driven power system safety assessment method and system, belonging to the field of power system safety assessment, wherein the power system safety assessment method comprises the following steps: collecting a multi-time scale data set of a power grid dispatching platform; predicting the day-ahead load of the power system according to the historical load data to obtain day-ahead load prediction data; carrying out load flow calculation on the power system according to the multi-time scale data set and the load prediction data before the day; determining a day-ahead safety evaluation index value of the power system based on a safety evaluation model according to the day-ahead fault type, the day-ahead power grid topological structure and the day-ahead power flow calculation result; carrying out load flow calculation on the power system according to the real-time load; and determining a real-time safety evaluation index value of the power system based on the safety evaluation model according to the current fault type, the current power grid topological structure and the real-time load flow calculation result. The efficiency of electric power system safety assessment is improved.
Description
Technical Field
The invention relates to the field of power system safety evaluation, in particular to a data-model hybrid-driven power system safety evaluation method and system.
Background
In the context of low-carbon transformation of energy structures, constructing a new power system based on new energy will become an important means to achieve the goal of "carbon peak-to-peak carbon neutralization", which makes the power system shift from a deterministic system to a strongly non-deterministic system. On the power supply side, high-proportion renewable energy becomes a main characteristic of a novel power system; unlike conventional hydroelectric power and thermal power, renewable energy power generation is influenced by meteorological conditions and environmental factors, and output of the renewable energy power generation shows the characteristics of intermittence and fluctuation; large-scale renewable energy access makes power system operation with significant uncertainty. On the load side, with the wide access of electric automobiles, the increasing frequency of supply and demand interaction and the development of photovoltaic and energy storage on the user side, the load shows the characteristics of activity and complexity. On the grid side, the transmission grid is affected by random source loads, and large-range tidal current fluctuation can occur. The influence of random disturbance or fault on the power system is mainly shown as branch overload and node voltage out-of-limit, which can cause the power grid to be impacted. The result of the random power flow can directly give the data of the branch power flow and the node voltage, but the system safety level cannot be judged only from the data of the two aspects. In order to ensure the safe and stable operation of the system, the establishment of a power grid analysis and calculation method, an operation safety evaluation method and related standards which are suitable for the new environment is increasingly urgent.
Massive new energy and novel power electronic equipment promote the construction of a novel power system, simultaneously greatly improve the complexity of system energy flow and information flow, lead to the sharp increase of the safety evaluation calculation time of the power system, and the influence on the scheduling of a large-scale actual power system cannot be ignored.
Disclosure of Invention
The invention aims to provide a safety assessment method and a safety assessment system for a data-model hybrid driven power system, which can improve the safety assessment efficiency of the power system.
In order to achieve the purpose, the invention provides the following scheme:
a safety assessment method for a data-model hybrid driven power system comprises the following steps:
collecting a multi-time scale data set of a power grid dispatching platform; the multi-time scale data set comprises planning data and daily occasional failure data; the planning data comprises a power grid topological structure under a monthly time scale, a unit on-off state under a weekly time scale and power grid operation data under a daily time scale; the daily and sporadic fault data comprise a power grid topological structure, power grid operation data and fault types when faults occur;
acquiring historical load data of a power system;
predicting the day-ahead load of the power system by adopting a Bayesian network according to the historical load data to obtain day-ahead load prediction data;
carrying out load flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead load flow calculation result; the day-ahead power flow calculation result comprises active power, reactive power and voltage distribution data of the power system;
acquiring a day-ahead fault type and a day-ahead power grid topological structure;
determining a day-ahead safety evaluation index value of the power system based on a safety evaluation model according to the day-ahead fault type, the day-ahead power grid topological structure and the day-ahead power flow calculation result; the security evaluation model is obtained by training XGboost by adopting a training sample set in advance; the training sample set comprises a plurality of groups of characteristic values and safety assessment index values corresponding to the characteristic values; each group of characteristic values comprises a historical fault type, a historical power grid topological structure and a historical load flow calculation result;
acquiring a real-time load of a power system;
carrying out load flow calculation on the power system according to the real-time load to obtain a real-time load flow calculation result;
acquiring a current fault type and a current power grid topological structure;
and determining a real-time safety evaluation index value of the power system based on the safety evaluation model according to the current fault type, the current power grid topological structure and the real-time power flow calculation result.
In order to achieve the purpose, the invention also provides the following scheme:
a data-model hybrid driven power system security assessment system, comprising:
the multi-scale data acquisition unit is used for acquiring a multi-time scale data set of the power grid dispatching platform; the multi-time scale data set comprises planning data and daily occasional failure data; the planning data comprises a power grid topological structure under a monthly time scale, a unit on-off state under a weekly time scale and power grid operation data under a daily time scale; the daily and sporadic fault data comprise a power grid topological structure, power grid operation data and fault types when faults occur;
a historical load acquisition unit for acquiring historical load data of the power system;
the day-ahead load prediction unit is connected with the historical load acquisition unit and used for predicting the day-ahead load of the power system by adopting a Bayesian network according to the historical load data to obtain day-ahead load prediction data;
the day-ahead load flow calculation unit is connected with the multi-scale data acquisition unit and the day-ahead load prediction unit and is used for carrying out load flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead load flow calculation result; the day-ahead power flow calculation result comprises active power, reactive power and voltage distribution data of the power system;
the day-ahead topology obtaining unit is used for obtaining the day-ahead fault type and the day-ahead power grid topology structure;
the day-ahead safety evaluation unit is connected with the day-ahead topology acquisition unit and the day-ahead power flow calculation unit and is used for determining day-ahead safety evaluation index values of the power system based on a safety evaluation model according to the day-ahead fault type, the day-ahead power grid topology structure and the day-ahead power flow calculation result; the security evaluation model is obtained by training XGboost by adopting a training sample set in advance; the training sample set comprises a plurality of groups of characteristic values and safety evaluation index values corresponding to the characteristic values; each group of characteristic values comprises a historical fault type, a historical power grid topological structure and a historical power flow calculation result;
the real-time load acquisition unit is used for acquiring the real-time load of the power system;
the real-time load flow calculation unit is connected with the real-time load acquisition unit and used for carrying out load flow calculation on the electric power system according to the real-time load to obtain a real-time load flow calculation result;
the current topology obtaining unit is used for obtaining a current fault type and a current power grid topology structure;
and the real-time safety evaluation unit is connected with the current topology acquisition unit, the real-time power flow calculation unit and the day-ahead safety evaluation unit and is used for determining a real-time safety evaluation index value of the power system based on the safety evaluation model according to the current fault type, the current power grid topology structure and the real-time power flow calculation result.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: collecting a multi-time scale data set of a power grid dispatching platform; predicting the day-ahead load of the power system by adopting a Bayesian network according to the historical load data to obtain day-ahead load prediction data; carrying out load flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead load flow calculation result; determining a day-ahead safety evaluation index value of the power system based on a pre-trained safety evaluation model according to a day-ahead fault type, a day-ahead power grid topological structure and a day-ahead power flow calculation result; and determining a real-time safety evaluation index value of the power system based on a pre-trained safety evaluation model according to the current fault type, the current power grid topological structure and the real-time load flow calculation result. Through an online and offline evaluation mode, the calculation burden is transferred to offline training, the safety evaluation of the power system in the day ahead and in real time is realized, and the safety evaluation efficiency is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a method for safety assessment of a power system driven by a data-model hybrid according to the present invention;
FIG. 2 is a block diagram of a data-model hybrid driven power system security assessment process;
FIG. 3 is a circuit diagram of a line fault handling;
FIG. 4 is a block diagram of a power system security assessment;
FIG. 5 is a training flow diagram of the XGboost model;
FIG. 6 is a block diagram of a safety evaluation system of a data-model hybrid driven power system according to the present invention.
Description of the symbols:
the system comprises a multi-scale data acquisition unit-1, a historical load acquisition unit-2, a day-ahead load prediction unit-3, a day-ahead load flow calculation unit-4, a day-ahead topology acquisition unit-5, a day-ahead safety evaluation unit-6, a real-time load acquisition unit-7, a real-time load flow calculation unit-8, a current topology acquisition unit-9 and a real-time safety evaluation unit-10.
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.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1 and fig. 2, the safety evaluation method for a data-model hybrid driven power system provided by the present invention includes:
s1: and collecting a multi-time scale data set of the power grid dispatching platform. The multi-time scale dataset includes planning data and occasional failure data. The planning data comprises a power grid topological structure under a monthly time scale, a unit on-off state under a weekly time scale and power grid operation data under a daily time scale. The daily and sporadic fault data comprise a power grid topological structure, power grid operation data and fault types when faults occur.
Specifically, on a monthly time scale, power grid topological structure data after lines and unit overhaul are collected from a power grid dispatching platform. And under the weekly time scale, the on-off state of the unit in the week is collected from the power grid dispatching platform. And acquiring information of power grid load, active power and reactive power of generator nodes and the like from the power grid dispatching platform on a day-time scale. When the power grid has accidental faults, the power grid topological structure, the on-off state of a unit and a line, the active power and the reactive power of a generator and a load and the like at the current moment are collected from the power grid dispatching platform in real time.
When the power grid fails, firstly, the failure type and the current power grid topological structure are confirmed,then calculating the probability of the fault occurrence: pr (total reflection) i ＝(d+1)/D s . In the formula D s D is the number of failures occurred so far, and d is the number of days from the date of normal operation of the power system to the current date.
S2: historical load data of the power system is obtained.
S3: and predicting the day-ahead load of the power system by adopting a Bayesian network according to the historical load data to obtain day-ahead load prediction data.
S4: and carrying out load flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead load flow calculation result. The day-ahead power flow calculation result comprises active power, reactive power and voltage distribution data of the power system.
Specifically, a random power flow model is established according to the multi-time scale data set and the day-ahead load prediction data; and solving the random power flow model by adopting a semi-invariant method, and determining active power, reactive power and voltage distribution data of the power system.
S5: and acquiring the day-ahead fault type and the day-ahead power grid topological structure.
S6: and determining a day-ahead safety evaluation index value of the power system based on a safety evaluation model according to the day-ahead fault type, the day-ahead power grid topological structure and the day-ahead power flow calculation result. The security evaluation model is obtained by adopting a training sample set to train the XGboost in advance. The training sample set comprises a plurality of groups of characteristic values and safety assessment index values corresponding to the characteristic values. Each group of characteristic values comprises historical fault types, historical power grid topological structures and historical load flow calculation results.
S7: and acquiring the real-time load of the power system.
S8: and carrying out load flow calculation on the power system according to the real-time load to obtain a real-time load flow calculation result.
S9: and acquiring the current fault type and the current power grid topological structure.
S10: and determining a real-time safety evaluation index value of the power system based on the safety evaluation model according to the current fault type, the current power grid topological structure and the real-time power flow calculation result.
Furthermore, in step S3, a Bayesian neural network-based day-ahead load prediction function considering uncertainty is realized. The Bayesian network consists of a statistical model, a neural network, and prior probability and likelihood probability. The method for predicting the day-ahead prediction by adopting the Bayesian network comprises the following steps:
(1) And (5) building a network model. Setting model training set I, I x For characteristic values, i.e. historical load, I y Is a predicted value. W represents neural network parameters, each network weight W i E W satisfies the mean value μ i Variance of δ i And each weight is independent, the prior probability p (W) of the weight parameter is given by the statistical model.
And (3) calculating the posterior probability based on Bayes theorem:
where p (W | I) represents the posterior probability of the network parameter, p (I) y |I x W) is at weight W, sample I x Lower, I y Can reflect the network estimation performance.
Distributing variation q by variation inference φ = q (W | θ) approximates p (W | I). First calculating KL divergence to obtain D KL (q | | p) for measuring different probability distribution distances. Then, a function ELBO is further calculated, ELBO maximization is achieved through a backprop algorithm, and p (W | I) is obtained indirectly.
Wherein, P (I) is the prior probability of the data, and is obtained by a probability statistical method based on historical load data.
(2) Constructing a Bayesian neural network loss function L (I) x,j )。
Wherein n represents the number of network weights, epsilon, gamma, alpha and beta are all preset hyper-parameters, and theta i ＝(μ i ,δ i )，
Is the prediction result corresponding to the jth sample, p (w) i ) Is a weight parameter w i A priori probability of.
(3) And (5) training and evaluating the model. And carrying out Bayesian neural network model training based on historical data. And (3) evaluating the model training result by selecting a classical prediction evaluation index, wherein the index comprises an average absolute error MAE, a mean square error MSE, a root mean square error RMSE and an R2_ score, and y i The actual value is represented by the value of,
represents the predicted value, and m is the number of samples.
When load flow calculation is performed in the step S4, the fixed basic data is derived from the multi-time scale data set, and the random data is derived from the day-ahead load prediction data.
In this embodiment, only considering that the injected power of each node is independent, the random power flow calculation steps are as follows:
(1) Raw data including line, generator, node related data is input.
(2) Information about load and random quantity of the generator is determined, normal distribution is adopted, and an expected value, variance and the like need to be determined.
(3) And determining the information of the fault line, including the fault line number and the fault rate.
(4) And calculating the power flow by using a Newton-Raphson method. Calculating the load flow information of the normal operation state, namely the node injection amount W 0 (including active and reactive power injected by each node), state variable X 0 (i.e., information about voltage amplitude and voltage angle of each node), branch flow variable Z 0 (including branch active power P ij And branch reactive power Q ij )。
(5) From Newton-Raphson power flow results, i.e. the state variable X 0 To find the Jacobian matrix J used for the last iteration of the Newton-Raphson method 0 Partial derivative matrix G of sum branch power flow equation 0 . According to the Jacobian matrix J 0 Determining a first sensitivity matrix S 0 (J 0 Inverse matrix of) and a second sensitivity matrix T 0 。
The branch flow equation is:
wherein, P ij Active power, Q, for branch ij ij Reactive power, V, for branch ij i Is the voltage amplitude of node i, G ij For the conductance of the line between node i and node j in the power network, θ ij For the voltage phase difference between node i and node j in the power network, B ij Is the inductance of the line between node i and node j in the power network, t ij Is the transformation ratio of the transformer branch, b ij0 Half that accommodated by branch ij.
The above formula can be expressed as Z = g (X), and the branch load flow Z = Z 0 + Δ Z, linearizing the branch flow equation to obtain:
ΔZ＝G 0 ΔX；
in the formula, Δ Z is the random disturbance of the branch power flow Z.
G 0 Element in (2) and Jacobian matrix J 0 The following relationships exist for the elements in (1):
in the formula, V i Is the voltage amplitude of node i, V j Is the voltage amplitude of the node, V k The voltage amplitude of node k, k ≠ i, j, θ i Is the voltage phase angle, θ, of node i j Is the phase angle of the voltage at node j, θ k Is the phase angle of the voltage at node k, H ij 、N ij 、J ij 、L ij Is an intermediate variable, θ ij Is the voltage phase difference between node i and node j in the power system.
Linearizing the node power equation to obtain delta W ≡ J 0 Δ X, where Δ W is the random perturbation of the node injection amount W, Δ X is the random perturbation of the state variable X, W = W 0 +ΔW，X＝X 0 +ΔX。
(6) Setting the load distribution as normal distribution, and obtaining semi-invariant gamma of each stage of load wL (k) 。
The new energy output is also obeyed normal distribution, and each order semi-invariant gamma of the new energy output is obtained wg (k) 。
Wherein, mu L Mean value of load, μ g Rated power, sigma, for new energy output L Is the standard deviation, σ, of the load distribution g And k represents the order of the semi-invariant, which is the standard deviation of the new energy output.
The handling of a line fault is a discrete distribution, as shown in FIG. 3, where Z g Is the impedance of the line, b 0 Half that accommodated by branch ij. And adding a virtual injection source to each node on two sides of the fault branch, wherein when the injected power is respectively equal to the power flowing out from two ends of the branch ij, the state of the system is the same as the state of the branch ij after disconnection.
Determining Δ P for the virtual implant source according to the following equation i 、ΔQ i 、ΔP j 、ΔQ j ：
Wherein, T 4×4 Is the second sensitivity matrix T 0 A sub-matrix of the first and second sensitivity matrices T, extracting a second sensitivity matrix T 0 The element of the corresponding branch ij in (1) constitutes T 4×4 . Line fault equivalent to Δ P i 、ΔQ i 、ΔP j 、ΔQ j Two power supply distribution of (1) 4×4 Is an identity matrix.
(7) The line fault is equivalent to binomial distribution, and the central moments of each order of the binomial distribution can be obtained
q i Is availability, C i Δ P as a virtual implant source i 、ΔQ i And the like.
By the central moments alpha of the orders v With semi-invariants of each order gamma v To find the semi-invariant of each step
γ 1 ＝α 1 ＝μ；
(8) And obtaining each-order semi-invariant of the delta W according to the additivity of the semi-invariant.
According to the theorem that the semi-invariant of each order of the sum of the independent random variables is equal to the semi-invariant of each order of each variable, the semi-invariant of each order of the delta W can be obtained
Equal to each order of semi-invariant of the generator injected power
And each step of semi-invariant of load injection power
And (4) summing.
When considering line faults, adding a semi-invariant quantity when considering the line faults
According to the theorem' random variable alpha times k order semi-invariant is equal to the variable k order semi-invariant alpha k Multiple ", can get:
in the formula (I), the compound is shown in the specification,
and
respectively a first sensitivity matrix S 0 A second sensitivity matrix T 0 Of the k-th power of the elements in (A), gamma x (k) And gamma z (k) Respectively, Δ X and Δ Z, respectively.
(9) Determining the central moments beta of the respective orders of DeltaX and DeltaZ from the relationship between the semi-invariants and the central moments v 。
β 1 ＝0；
β 2 ＝γ 2 ＝σ 2 ；
β 3 ＝γ 3 ；
β 5 ＝γ 5 +10γ 3 γ 2 ；
(10) Calculating the coefficient c according to the relation between the central moment and the coefficient of Gram-Charlie expansion series v 。
c 0 ＝1；
c 1 ＝0；
c 2 ＝0；
(11) Coefficient c according to Gram-Charlie expansion series v Determining the probability density function and the cumulative distribution function of Δ X and Δ Z:
the probability density function and the cumulative distribution function for the random variable x are expressed as Gram-Charlier expansion series as follows.
F(x)＝Φ(x)+c 1 Φ'(x)+c 2 Φ”(x)+c 3 Φ”'(x)+…；
Where F (x) is the probability distribution function of the random variable x, and F (x) is the probability density function of x. Phi (x) and
a distribution function and a probability density function of a normal distribution of the desired m =0 and standard deviation σ =1, respectively.
The Gram-Charlier series expansion is determined using a Hermite polynomial.
H 0 (x)＝1；
H 1 (x)＝-x；
H 2 (x)＝x 2 -1；
H 3 (x)＝(x 3 -3x)×(-1)；
H 4 (x)＝x 4 -6x+ 3；
H 5 (x)＝(x 5 -10x 3 +15x)×(-1)；
H 6 (x)＝x 6 -15x 4 +45x 2 -15。
From this, the Gram-Charlier series expansion of the random variable distribution function can be obtained:
in the formula (I), the compound is shown in the specification,
is a variable normalized by a random variable x,
after the random variables Δ X and Δ Z are normalized, the probability density function and the cumulative distribution function of Δ X and Δ Z are obtained. Because the state variable X = X 0 + Δ X and branch current Z = Z 0 + Δ Z, so the probability density function and cumulative distribution function of Δ X are shifted by X 0 Obtaining the probability density function and the cumulative distribution function of X, translating the probability density function and the cumulative distribution function of Delta Z by Z 0 The unit obtains the probability density function and the cumulative distribution function of Z.
The data driving method is characterized in that the output uncertainty variable is completely deduced from historical data, a certain fixed probability density function type is not used any more, and the range of the probability density function is given by calculating the variables such as the mean value, the variance and the like of the historical data. Therefore, the distribution of uncertainty can be more accurately plotted.
In this embodiment, a deviation degree measurement method is adopted to establish element-level severity indexes, which are the branch load flow out-of-limit severity and the node voltage out-of-limit severity, so as to evaluate the branch load flow out-of-limit and node voltage out-of-limit consequences.
The training process of the safety assessment model comprises the following steps:
s101: and acquiring a plurality of groups of characteristic values. Each group of characteristic values comprises a historical fault type, a historical power grid topological structure and a historical load flow calculation result; the historical power grid topological structure comprises a plurality of nodes and a plurality of branches.
S102: and aiming at any group of characteristic values, determining an expected fault set according to the historical fault type and the historical power grid topological structure. The set of expected failures includes a plurality of expected failures.
The invention establishes the element probability model from two aspects of the generator and the transmission line respectively. The generator probability model is a probability model of the generator in a normal operation state or an accident state, and is divided into two types according to whether time interval influence is considered or not: a state probability model within a single time period and a multi-period state transition probability model that takes into account timing. Within a single time period, there are generally two types of probabilistic models for a generator: a two-state model and a derated state model.
The two-state model means that the generator only has two states of rated operation and shutdown, and when the model is used for non-sequential Monte Carlo simulation, the operation state of the generator is assumed to be S i ，PF i For failure probability, the generator i is generated in [0,1 ]]Random number R with uniformly distributed intervals i :
The derated state model is also referred to as a multi-state model. During multi-state Monte Carlo analog sampling, the generator i is generated at [0, 1%]Random number R with uniformly distributed intervals i Carrying out simulation:
when considering a plurality of continuous periods, the running state of the generator can change along with the scheduling condition of each period, and the running state can be represented by a state transition model, wherein each state transition parameter respectively represents the transition probability of the generator among running, outage and derating states, and is obtained based on historical data or short-term prediction data fitting.
Counting the fault rate lambda (times/year) and the repair rate mu (times/year) of the transmission line based on historical data, calculating the forced outage rate FOR = lambda/(lambda + mu), extracting (0, 1) uniformly distributed random numbers R, and considering the line fault if R is less than or equal to FOR. The proportion of open circuit and short circuit fault in the fault is P 0 And P S And sampling R again, and further judging the fault type: when R is<P S When the line is open, the line is considered to be broken.
Further considering the influence of weather factors on the fault probability of the power transmission line, and obtaining the fault rate lambda of the whole line in severe climate and normal climate according to the series connection characteristic of the lines 1e Repairing time r 1e And an unavailability rate U 1e ：
λ 1e ＝λ ad R+λ no (1-R)；
U 1e ＝U ad +U no -U ad U no ；
Wherein λ is ad For failure rate under severe weather conditions, R is the percentage value of the line length exposed to severe weather conditions, lambda no Failure rate under normal climatic conditions, r ad Repair time in bad weather conditions, r no Repair time under normal climatic conditions, U ad Is the unavailability rate in severe weather conditions, U no Is the unavailable rate under normal climatic conditions.
And forming an expected fault set based on the generator probability model and the transmission line probability model, and calculating the probability that the load flow of each branch circuit is not out of limit and the voltage of each node is not out of limit under various fault conditions.
The expected failure set E is:
wherein the content of the first and second substances,
for generator i in off-stream or derated state event, E line And (3) calculating the probability of each fault occurrence for the power transmission line fault event by using the formula in the step (S1).
In this embodiment, when the whole period of the probabilistic safety analysis is sufficiently small (e.g. less than 15 minutes), the set of expected failures is determined according to the N-1 method:
wherein prob {. Is an event occurrence probability function; f pq (. Is a probability distribution function of active and reactive power of the lower branch power flow of the fault E, F vθ (. Is a probability distribution function of the node voltage amplitude and phase angle, σ pj Is the standard deviation of the power flow of branch j, sigma pj Is the standard deviation of the voltage amplitude of node i.
S103: and determining the power flow out-of-limit severity of each branch and the voltage out-of-limit severity of each node according to the expected fault set, the historical power grid topological structure and the historical power flow calculation result.
The branch current threshold severity represents the percentage of the current of each line to the delivery capacity limit. Defined as the degree of out-of-limit of the branch ij current flow for all K possible fault cases. Determining the out-of-limit severity of the power flow of the branch ij by adopting the following formula:
wherein the content of the first and second substances,
is the out-of-limit severity of the flow for branch ij,
in order to cause the fault set of the branch ij with the power flow out of limit, E is an expected fault set, K is the total number of the expected faults,
for the active power of branch ij in anticipation of the occurrence of fault k, P max The maximum active power value allowed to be transmitted for branch ij is obtained from the thermally stable limiting current, P min For the minimum value of active power allowed to be transmitted by branch ij, | | is the absolute value.
The node voltage out-of-limit severity represents the percentage of each node voltage amplitude that deviates from the normal amplitude limit, and is defined as the out-of-limit degree of the node i voltage under all possible K fault conditions. Determining the voltage out-of-limit severity of node i using the following equation:
wherein the content of the first and second substances,
is the voltage of node i out-of-limit severity,
to cause the voltage amplitude of the node i to exceed the limit, E is an expected fault event set, K is the total number of expected faults, v H For the maximum voltage amplitude allowed at node i, if the maximum voltage amplitude constraint is exceeded it will cause the voltage to collapse,
when a failure k is expected for a node iVoltage amplitude of v L If the minimum voltage amplitude allowed for the node i is smaller than the minimum voltage amplitude constraint, the low voltage instability of the system is caused.
S104: and determining the overload severity index value of each branch according to the out-of-limit severity of the tide of each branch.
The overload risk index of the power system reflects the average level of overload risks of each branch and considers the influence of the maximum risk value. Determining the overload severity index value of the branch by adopting the following formula:
wherein R is OL Alpha and beta are weight coefficients, alpha + beta =1, M is the total number of branches in the power system, sev is the branch overload severity index value branch A matrix formed of the severity of the tidal current violation of each branch, sev branch || 1 Pair of expression Sev branch Calculating 1 norm, | | Sev branch || ∞ Pair of representation branch And calculating an infinite norm.
S105: and determining the node out-of-limit severity index value according to the voltage out-of-limit severity of each node.
The out-of-limit severity index of the power system node reflects the average level of the out-of-limit severity of the voltage amplitude of each node, and considers the influence of element fault events which have the largest influence on the system. Determining a node out-of-limit severity index value by adopting the following formula:
wherein R is OV The node out-of-limit severity index value is defined, alpha and beta are weight coefficients, alpha + beta =1, N is the total number of nodes in the power system, sev voltage Is a matrix of voltage off-limit severity of each node, | Sev voltage || 1 Pair of expression Sev voltage Calculating 1 norm, | | Sev voltage || ∞ Pair of expression Sev voltage And calculating an infinite norm.
S106: the probability of occurrence of each expected fault, the safety probability of the power system when each expected fault occurs, and the safety probability of the power system when no fault occurs are calculated.
The safety probability of the power system is the probability that branch flow overload or node voltage out-of-limit occurs to the power system is calculated for all expected faults. In anticipation of Accident E i After the occurrence, the probability that the power system does not satisfy the safety constraint is:
consider hypothetical event E i Obey poisson distribution, and each event is independent, then the probability of each fault occurrence is:
for each expected failure E k (k =1,2,. Multidot.N) calculating the probability Pr of the system meeting the power flow safety constraint s1 ,Pr s2 ,…,Pr sN And the system meets the tidal current safety constraint under the condition of no fault to obtain the probability Pr s0 。
S107: and determining an unsafe probability index value according to the probability of occurrence of each expected fault, the safety probability of the power system when each expected fault occurs and the safety probability of the power system when no fault occurs.
Specifically, the unsafe probability index value is calculated using the following formula:
wherein, pr ins Pr is an index value of the probability of insecurity s0 Is the safety probability of the power system without faults, K is the total number of expected faults,
to anticipate failure E k Probability of occurrence, pr sk To anticipate failure E k Safety probability of the power system at the time of occurrence.
S108: and determining a safety evaluation index value corresponding to the group of characteristic values according to the branch overload severity index value, the node out-of-limit severity index value and the unsafe probability index value.
Specifically, the following formula is adopted to determine the safety evaluation index value:
S 0 ＝w 1 ·Pr ins +w 2 ·R OL +w 3 ·R OV ；
wherein S is 0 For the safety evaluation of the index value, w 1 Weight of the unsafe probability index, w 2 Weight, w, for branch overload severity indicator 3 Weight of node out-of-limit severity indicator, pr ins As an index value of the probability of insecurity, R OL Is a branch overload severity index value, R OV And the index value is the node out-of-limit severity index value. The weight can be determined by subjective and objective weighting and the like.
S109: and training the XGboost according to each group of characteristic values and the security evaluation index value of each group of characteristic values to obtain a security evaluation model.
The safety evaluation model constructed by the invention comprises a Bayesian neural network for day-ahead randomness prediction on one hand and an XGboost network for safety evaluation on the other hand. In the future, according to the probability density function of the Bayesian neural network for wind and light load randomness prediction, the XGboost input is finally formed by combining other boundary conditions provided by the scheduling platform, and the evaluation on the system security of the next day is realized. On one hand, the change of boundary conditions such as network architecture and installed capacity can dynamically update two networks according to the input of a scheduling platform and the statistics of actual load flow conditions, so that the self-adaptive updating of the two neural network parameters is realized; on the other hand, in the event of a fault, the weighting parameters of the XGboost may be updated according to the type of fault and the frequency of the contingency.
In order to improve the accuracy and speed of security prediction, before the XGBoost is trained, the feature values in the training sample set are firstly: and (4) performing z-score standardization on historical power flow calculation results (branch active power flow distribution, mean value and variance of voltage amplitude values and the like).
Wherein x' is normalized data, x is data before normalization, x μ Mean of data before normalization, x σ Is the standard deviation of the data before normalization.
All the collected data are respectively standardized according to the types by adopting the formula, and the variance and the mean value are specific to a certain type of data. For example, the standardization process of the active power flow variance in the sample is as follows: and calculating the variance and the mean of the data such as the active power flow variance, and standardizing the data.
The safety evaluation of the power system comprises two parts of day-ahead evaluation and real-time evaluation, and on-line reasoning is completed on the basis of off-line training, so that the evaluation efficiency of the safety evaluation is improved.
In the safety evaluation in the day ahead, the fault type and the power grid topological structure of multi-time scale data are input into the model in the off-line training process, and the historical power flow, the voltage distribution mean value and the voltage distribution variance data of the random power flow calculation result are obtained. And in the online reasoning process, the safety assessment model inputs the fault type, the power grid topological structure, newly calculated power flow calculated based on the day-ahead predicted load, the voltage distribution mean value and the voltage distribution variance, and outputs a day-ahead safety assessment result.
In the real-time safety evaluation function, the model inputs the fault type and the power grid topological structure of multi-time scale data in the off-line training process, and the model historical load flow, the voltage distribution mean value and the voltage distribution variance data of the random load flow calculation result. And in the online reasoning process, the safety assessment model inputs the fault type, the power grid topological structure, the power flow calculated based on the real-time load, the voltage distribution mean value and the voltage distribution variance, and outputs a real-time safety assessment result.
Specifically, when a security assessment model is established based on XGboost, a system security assessment scene is designed first, and a security assessment training sample set is constructed based on data required by the day-ahead security assessment and the real-time security assessment. Constructing a sample set X = [ X ] representing random power flow input and output changes in a normal operation state and a fault state based on the corresponding fault set and data acquisition points 1 ,...,X i ,...,X N ]∈R N×M . Wherein, N represents the number of samples, namely N security assessment scenes, and M represents the dimensionality of each sample feature vector. Sample X i The input feature vector comprises a mean value { f of branch active power flow distribution h L H is in the middle of H and variance { sigma h | H ∈ H }, and the magnitude of the node voltage { v ∈ H }, and a la ∈ A } and variance { ε a The dimension number M is the sum of the branch number H and the node number A. Each sample X i Corresponding probability of system insecurity y i Constituting the output Y of the security assessment model.
And constructing a safety evaluation model based on the training sample set. The system security assessment framework as shown in fig. 4, for a given dataset D = { (X) i ,y i )}(|D|＝N,X i ∈R M ,y i E R), the integration model of the tree is represented by:
in the formula (I), the compound is shown in the specification,
is the output result of the model; t is a unit of k (X i ) Representing the prediction result of the kth decision tree;
representing a mapping of node input samples to leaf nodes; each tree T k Weights corresponding to an independent tree structure q and leaf
X i Is the feature vector of the ith sample; q represents the index of the mapping of the structure of each tree to the leaf corresponding to the sample; Λ is the number of leaves on the tree;
is the collection space of the tree.
The XGboost trained objective function is:
in the formula (I), the compound is shown in the specification,
is a model parameter;
the quantization error of the model on the training samples, and N is the number of the training samples;
and K is the number of model base learners. C is a constant. Gamma-shaped i (X i ) Representing a mapping of the determined input sample values to leaf nodes.
The model complexity of a single base learner in the XGboost algorithm is as follows:
in the formula, M is a base learning device T k The leaf node number of (1), lambda represents the L2 regularization coefficient, gamma represents the node segmentation difficulty,
the L2 norm representing the leaf node weight.
Under the general condition, the XGboost model is trained by adopting an incremental training method, namely, the XGboost model is trained on the basis of keeping the original model each timeAdding a new function (i.e. a new tree) to the model achieves better performance by integrating a series of less-learning base learners.
Smaller values of (c) indicate better structure of the tree. The increment function added in each round reduces the target function to the maximum extent as possible, and the kth base learning device T is trained k The learning objective function of time is:
in the formula (I), the compound is shown in the specification,
is T k The parameters of (1); omega (T) k ) Is T k The model complexity of (2);
model residual error of previous iteration;
is T k An output of (d); the learning rate ε is a range (0, 1).
And continuously training the CART-based learner to fit the residual error of the previous model based on the training sample set and integrating the residual error into the XGboost model until a preset number of the CART-based learners are trained or the residual error of the model is smaller than a set threshold value.
In the formula (I), the compound is shown in the specification,
is the firstModel predicted values of i samples at the t-th round,
model prediction value of retaining t-1 round
Then, a new function T is added t (X i ). The training flow chart is shown in fig. 5.
The method realizes the safety evaluation suitable for the novel power system by combining online and offline model-data driven evaluation. Firstly, a multi-time scale novel power system safety assessment data set covering month, week and day power grid framework and operation data is constructed by collecting power grid dispatching platform data on line; then establishing a random power flow model of the new energy power system and solving the random power flow model based on a semi-invariant method; and finally, the data are dynamically interacted with a power grid dispatching platform to update data and train a data driving model in real time, so that a complex calculation process is converted into matrix operation, the calculation burden is transferred to off-line training, the safety assessment of the multi-scale new energy power system in the day ahead and in real time is realized, and the safety assessment calculation efficiency of the new energy power system is improved. The system comprises a multi-time scale data module, a random power flow calculation module and a multi-time scale new energy power system safety evaluation module based on data driving.
The method can directly derive data such as annual inter-provincial power transmission line power flow, monthly maintenance plan, weekly unit combination plan, day-ahead market clearing result and the like from the scheduling platform. On one hand, the data form a system operation boundary condition in the random power flow calculation and provide a probability density function of uncertain sources (new energy power generation, user power load and line outage) in the random power flow calculation; on the other hand, in combination with the output of the random power flow module, the data can also be used as the input of the XGboost network, and is used for training the network and evaluating the system security of a typical day under different time scales. The calculation process of the static security assessment of the traditional power system is simplified, the method is suitable for a novel power system security assessment scene with new energy access, the security assessment speed is high, and the robustness is strong.
As shown in fig. 6, the safety evaluation system for a data-model hybrid driven power system of the present invention includes: the system comprises a multi-scale data acquisition unit 1, a historical load acquisition unit 2, a day-ahead load prediction unit 3, a day-ahead power flow calculation unit 4, a day-ahead topology acquisition unit 5, a day-ahead safety evaluation unit 6, a real-time load acquisition unit 7, a real-time power flow calculation unit 8, a current topology acquisition unit 9 and a real-time safety evaluation unit 10.
The multi-scale data acquisition unit 1 is used for acquiring a multi-time scale data set of the power grid dispatching platform. The multi-time scale data set comprises planning data and daily and sporadic fault data; the planning data comprises a power grid topological structure under a monthly time scale, a unit on-off state under a weekly time scale and power grid operation data under a daily time scale; the daily and sporadic fault data comprise a power grid topological structure, power grid operation data and fault types when faults occur.
The historical load acquisition unit 2 is used for acquiring historical load data of the power system.
The day-ahead load prediction unit 3 is connected with the historical load acquisition unit 2, and the day-ahead load prediction unit 3 is used for predicting the day-ahead load of the power system by adopting a Bayesian network according to the historical load data to obtain day-ahead load prediction data.
The day-ahead power flow calculation unit 4 is connected with the multi-scale data acquisition unit 1 and the day-ahead load prediction unit 3, and the day-ahead power flow calculation unit 4 is used for performing power flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead power flow calculation result. The day-ahead power flow calculation result comprises active power, reactive power and voltage distribution data of the power system.
The day-ahead topology obtaining unit 5 is used for obtaining day-ahead fault types and day-ahead power grid topology structures.
The day-ahead safety evaluation unit 6 is connected with the day-ahead topology obtaining unit 5 and the day-ahead power flow calculating unit 4, and the day-ahead safety evaluation unit 6 is used for determining a day-ahead safety evaluation index value of the power system based on a safety evaluation model according to the day-ahead fault type, the day-ahead power grid topology structure and the day-ahead power flow calculation result; the security evaluation model is obtained by adopting a training sample set to train the XGboost in advance; the training sample set comprises a plurality of groups of characteristic values and safety evaluation index values corresponding to the characteristic values; each group of characteristic values comprises historical fault types, historical power grid topological structures and historical load flow calculation results.
The real-time load obtaining unit 7 is configured to obtain a real-time load of the power system.
The real-time load flow calculation unit 8 is connected with the real-time load acquisition unit 7, and the real-time load flow calculation unit 8 is used for performing load flow calculation on the power system according to the real-time load to obtain a real-time load flow calculation result.
The current topology obtaining unit 9 is configured to obtain a current fault type and a current power grid topology structure.
The real-time safety evaluation unit 10 is connected with the current topology obtaining unit 9, the real-time power flow calculation unit 8 and the day-ahead safety evaluation unit 6, and the real-time safety evaluation unit 10 is configured to determine a real-time safety evaluation index value of the power system based on the safety evaluation model according to the current fault type, the current power grid topology structure and the real-time power flow calculation result.
Compared with the prior art, the data-model hybrid-driven power system safety assessment system provided by the invention has the same beneficial effects as the aforementioned data-model hybrid-driven power system safety assessment method, and is not repeated herein.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (10)
1. A safety assessment method for a data-model hybrid driven power system is characterized by comprising the following steps:
collecting a multi-time scale data set of a power grid dispatching platform; the multi-time scale data set comprises planning data and daily occasional failure data; the planning data comprises a power grid topological structure under a monthly time scale, a unit on-off state under a weekly time scale and power grid operation data under a daily time scale; the daily and sporadic fault data comprise a power grid topological structure, power grid operation data and fault types when faults occur;
acquiring historical load data of a power system;
predicting the day-ahead load of the power system by adopting a Bayesian network according to the historical load data to obtain day-ahead load prediction data;
carrying out load flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead load flow calculation result; the day-ahead power flow calculation result comprises active power, reactive power and voltage distribution data of the power system;
acquiring a day-ahead fault type and a day-ahead power grid topological structure;
determining a day-ahead safety evaluation index value of the power system based on a safety evaluation model according to the day-ahead fault type, the day-ahead power grid topological structure and the day-ahead power flow calculation result; the security evaluation model is obtained by adopting a training sample set to train the XGboost in advance; the training sample set comprises a plurality of groups of characteristic values and safety assessment index values corresponding to the characteristic values; each group of characteristic values comprises a historical fault type, a historical power grid topological structure and a historical load flow calculation result;
acquiring a real-time load of a power system;
carrying out load flow calculation on the power system according to the real-time load to obtain a real-time load flow calculation result;
acquiring a current fault type and a current power grid topological structure;
and determining a real-time safety evaluation index value of the power system based on the safety evaluation model according to the current fault type, the current power grid topological structure and the real-time power flow calculation result.
2. The data-model hybrid driven power system safety assessment method according to claim 1, wherein the performing a power flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead power flow calculation result specifically comprises:
establishing a random power flow model according to the multi-time scale data set and the day-ahead load prediction data;
and solving the random power flow model by adopting a semi-invariant method, and determining active power, reactive power and voltage distribution data of the power system.
3. The data-model hybrid driven power system safety assessment method according to claim 1, wherein the training process of the safety assessment model comprises:
acquiring a plurality of groups of characteristic values; each group of characteristic values comprises a historical fault type, a historical power grid topological structure and a historical power flow calculation result; the historical power grid topological structure comprises a plurality of nodes and a plurality of branches;
aiming at any group of characteristic values, determining an expected fault set according to the historical fault type and the historical power grid topological structure; the set of expected faults includes a plurality of expected faults;
determining the power flow out-of-limit severity of each branch and the voltage out-of-limit severity of each node according to the expected fault set, the historical power grid topological structure and the historical power flow calculation result;
determining branch overload severity index values according to the load flow out-of-limit severity of each branch;
determining node out-of-limit severity index values according to the voltage out-of-limit severity of each node;
calculating the probability of occurrence of each expected fault, the safety probability of the power system when each expected fault occurs and the safety probability of the power system when no fault occurs;
determining an unsafe probability index value according to the probability of occurrence of each expected fault, the safety probability of the power system when each expected fault occurs and the safety probability of the power system when no fault occurs;
determining a safety evaluation index value corresponding to the group of characteristic values according to the branch overload severity index value, the node out-of-limit severity index value and the unsafe probability index value;
and training the XGboost according to each group of characteristic values and the safety evaluation index value of each group of characteristic values to obtain a safety evaluation model.
4. The data-model hybrid driven power system safety assessment method according to claim 3, characterized in that the load flow out-of-limit severity of branch ij is determined using the following formula:
wherein, the first and the second end of the pipe are connected with each other,
for the out-of-limit severity of the flow for branch ij,
in order to cause the fault set of the branch ij with the power flow out of limit, E is an expected fault set, K is the total number of expected faults,
for branch ij in advanceActive power when thinking of the occurrence of the fault k, P max Maximum value of active power, P, allowed to be transmitted for branch ij min For the minimum active power value allowed to be transmitted by branch ij, | | is the absolute value.
5. The data-model hybrid driven power system safety assessment method according to claim 3, characterized in that the voltage violation severity of node i is determined using the following formula:
wherein the content of the first and second substances,
is the out-of-limit severity of the voltage at node i,
in order to cause the fault set of the out-of-limit voltage amplitude of the node i, E is an expected fault event set, K is the total number of expected faults, v H The maximum voltage amplitude allowed for node i,
is the voltage amplitude, v, of node i in anticipation of the occurrence of fault k L The minimum voltage magnitude allowed for node i.
6. The data-model hybrid driven power system safety assessment method according to claim 3, wherein the branch overload severity index value is determined using the following formula:
wherein R is OL The index value of overload severity of branch is shown, alpha and beta are weight coefficients, alpha + beta =1, M is shown in the power systemTotal number of branches, sev branch A matrix formed by the out-of-limit severity of the tide of each branch, | | Sev branch || 1 Pair of expression Sev branch Calculating 1 norm, | | Sev branch || ∞ Pair of representation branch And calculating an infinite norm.
7. The data-model hybrid driven power system safety assessment method according to claim 3, wherein the node violation severity index value is determined using the following formula:
wherein R is OV The node out-of-limit severity index value is defined, alpha and beta are weight coefficients, alpha + beta =1, N is the total number of nodes in the power system, sev voltage Is a matrix of voltage off-limit severity of each node, | Sev voltage || 1 Pair of representation voltage Calculating 1 norm, | | Sev voltage || ∞ Pair of representation voltage And calculating an infinite norm.
8. The data-model hybrid driven power system safety assessment method according to claim 3, wherein the unsafe probability index value is calculated using the following formula:
wherein, pr ins Pr is an index value of the unsafe probability s0 Is the safety probability of the power system without faults, K is the total number of expected faults,
to anticipate failure E k Probability of occurrence, pr sk To anticipate failure E k Safety probability of the power system at the time of occurrence.
9. The data-model hybrid driven power system safety assessment method according to claim 3, wherein the following formula is adopted to determine the safety assessment index value:
S 0 ＝w 1 ·Pr ins +w 2 ·R OL +w 3 ·R OV ；
wherein S is 0 For evaluating the index value safely, w 1 Weight of the unsafe probability index, w 2 Weight, w, of the branch overload severity indicator 3 Weight of node out-of-limit severity indicator, pr ins As an index value of the probability of insecurity, R OL Is a branch overload severity index value, R OV And the index value is the out-of-limit severity index value of the node.
10. A safety assessment system for a data-model hybrid driven power system, the safety assessment system comprising:
the multi-scale data acquisition unit is used for acquiring a multi-time scale data set of the power grid dispatching platform; the multi-time scale data set comprises planning data and daily and sporadic fault data; the planning data comprises a power grid topological structure under a monthly time scale, a unit on-off state under a weekly time scale and power grid operation data under a daily time scale; the daily and sporadic fault data comprise a power grid topological structure, power grid operation data and fault types when faults occur;
a historical load acquisition unit for acquiring historical load data of the power system;
the day-ahead load prediction unit is connected with the historical load acquisition unit and used for predicting the day-ahead load of the power system by adopting a Bayesian network according to the historical load data to obtain day-ahead load prediction data;
the day-ahead load flow calculation unit is connected with the multi-scale data acquisition unit and the day-ahead load prediction unit and is used for carrying out load flow calculation on the power system according to the multi-time scale data set and the day-ahead load prediction data to obtain a day-ahead load flow calculation result; the day-ahead power flow calculation result comprises active power, reactive power and voltage distribution data of the power system;
the day-ahead topology obtaining unit is used for obtaining the day-ahead fault type and the day-ahead power grid topology structure;
the day-ahead safety evaluation unit is connected with the day-ahead topology acquisition unit and the day-ahead power flow calculation unit and is used for determining day-ahead safety evaluation index values of the power system based on a safety evaluation model according to the day-ahead fault type, the day-ahead power grid topology structure and the day-ahead power flow calculation result; the security evaluation model is obtained by adopting a training sample set to train the XGboost in advance; the training sample set comprises a plurality of groups of characteristic values and safety evaluation index values corresponding to the characteristic values; each group of characteristic values comprises a historical fault type, a historical power grid topological structure and a historical load flow calculation result;
the real-time load acquisition unit is used for acquiring a real-time load of the power system;
the real-time load flow calculation unit is connected with the real-time load acquisition unit and used for carrying out load flow calculation on the electric power system according to the real-time load to obtain a real-time load flow calculation result;
the current topology obtaining unit is used for obtaining a current fault type and a current power grid topology structure;
and the real-time safety evaluation unit is connected with the current topology acquisition unit, the real-time power flow calculation unit and the day-ahead safety evaluation unit and is used for determining a real-time safety evaluation index value of the power system based on the safety evaluation model according to the current fault type, the current power grid topology structure and the real-time power flow calculation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211018209.1A CN115392697A (en) | 2022-08-24 | 2022-08-24 | Data-model hybrid driven power system safety assessment method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211018209.1A CN115392697A (en) | 2022-08-24 | 2022-08-24 | Data-model hybrid driven power system safety assessment method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115392697A true CN115392697A (en) | 2022-11-25 |
Family
ID=84120475
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211018209.1A Pending CN115392697A (en) | 2022-08-24 | 2022-08-24 | Data-model hybrid driven power system safety assessment method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115392697A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115936448A (en) * | 2023-02-13 | 2023-04-07 | 南京深科博业电气股份有限公司 | Urban distribution network power evaluation system and method based on big data |
| CN116167527A (en) * | 2023-04-21 | 2023-05-26 | 南方电网数字电网研究院有限公司 | Pure data-driven power system static safety operation risk online assessment method |
-
2022
- 2022-08-24 CN CN202211018209.1A patent/CN115392697A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115936448A (en) * | 2023-02-13 | 2023-04-07 | 南京深科博业电气股份有限公司 | Urban distribution network power evaluation system and method based on big data |
| CN116167527A (en) * | 2023-04-21 | 2023-05-26 | 南方电网数字电网研究院有限公司 | Pure data-driven power system static safety operation risk online assessment method |
| CN116167527B (en) * | 2023-04-21 | 2023-09-12 | 南方电网数字电网研究院有限公司 | Pure data-driven power system static safety operation risk online assessment method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gao et al. | Day-ahead power forecasting in a large-scale photovoltaic plant based on weather classification using LSTM | |
| Rejc et al. | Short-term transmission-loss forecast for the slovenian transmission power system based on a fuzzy-logic decision approach | |
| CN115392697A (en) | Data-model hybrid driven power system safety assessment method and system | |
| CN113609649B (en) | Method for constructing medium-voltage line planning model of power distribution network based on opportunity constraint | |
| CN109088407B (en) | Power distribution network state estimation method based on deep belief network pseudo-measurement modeling | |
| CN110210670A (en) | A kind of prediction technique based on power-system short-term load | |
| CN111900713A (en) | Multi-scene power transmission network planning method considering load and wind power randomness under network source coordination | |
| CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
| CN114742283A (en) | Medium-voltage line loop closing current estimation and risk assessment method based on deep learning network | |
| CN113537582A (en) | Photovoltaic power ultra-short-term prediction method based on short-wave radiation correction | |
| CN108694479A (en) | Consider the distribution network reliability prediction technique that weather influences time between overhaul | |
| CN110490378A (en) | The calculation method of Power Network Status Estimation precision based on cloud SCADA big data | |
| Talayero et al. | Machine Learning models for the estimation of the production of large utility-scale photovoltaic plants | |
| Gao et al. | Evaluation on the short-term power supply capacity of an active distribution system based on multiple scenarios considering uncertainties | |
| Fan et al. | Ultra-short-term bus load forecasting method based on multi-source data and hybrid neural network | |
| CN115660893A (en) | Transformer substation bus load prediction method based on load characteristics | |
| CN115276067A (en) | Distributed energy storage voltage adjusting method adaptive to topological dynamic change of power distribution network | |
| Sun et al. | Reliability prediction of distribution network based on PCA-GA-BP neural network | |
| CN112232714B (en) | Deep learning-based risk assessment method for distribution network under incomplete structural parameters | |
| CN114429248A (en) | Transformer apparent power prediction method | |
| Wang et al. | An alternative method for estimating wind-power capacity credit based on reliability evaluation using intelligent search | |
| Ren et al. | A study of wind speed prediction based on particle swarm algorithm to optimize the parameters of sparse least squares support vector | |
| Singla et al. | Application of Levenberg Marquardt Algorithm for Short Term Load Forecasting: A theoretical investigation. | |
| Sičaja et al. | Evaluation of load forecast model performance in Croatian DSO | |
| Wu et al. | Short-term forecast of photovoltaic power generation output based on improved PSO-Elman neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |