CN113837432A - Power system frequency prediction method driven by physics-data combination - Google Patents
Power system frequency prediction method driven by physics-data combination Download PDFInfo
- Publication number
- CN113837432A CN113837432A CN202110922955.2A CN202110922955A CN113837432A CN 113837432 A CN113837432 A CN 113837432A CN 202110922955 A CN202110922955 A CN 202110922955A CN 113837432 A CN113837432 A CN 113837432A
- Authority
- CN
- China
- Prior art keywords
- power system
- frequency
- data
- prediction
- generator
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000012360 testing method Methods 0.000 claims abstract description 27
- 238000004088 simulation Methods 0.000 claims abstract description 17
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 230000001052 transient effect Effects 0.000 claims abstract description 5
- 238000011156 evaluation Methods 0.000 claims description 13
- 238000013016 damping Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 2
- 241000288105 Grus Species 0.000 claims 1
- 230000004044 response Effects 0.000 description 4
- 125000004122 cyclic group Chemical group 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Tourism & Hospitality (AREA)
- Medical Informatics (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Mathematical Physics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Primary Health Care (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a power system frequency prediction method driven by physics-data combination, belonging to the field of power system frequency prediction. The method comprises the following steps: performing electromechanical transient simulation on the power system by adopting simulation software to generate power system disturbance data; selecting a first part of input characteristic quantity and output quantity; calculating a second part of input characteristic quantity by using the selected power system frequency-related physical knowledge and the first part of input characteristic quantity; normalizing the input characteristic quantity and the output quantity of the two parts; dividing data into training samples and testing samples, constructing a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network, and inputting the testing samples into the trained frequency prediction model to realize rapid and accurate prediction of a power system frequency curve. The method provided by the invention can be used for predicting the frequency curve more accurately, and has stronger prediction generalization capability, anti-noise capability and interpretability of results.
Description
Technical Field
The invention belongs to the field of power system frequency prediction, and particularly relates to a power system frequency prediction method driven by physics-data in a combined mode.
Background
In recent years, with the large-scale grid connection of new energy units such as wind power and photovoltaic, the installed proportion of the traditional synchronous unit is gradually reduced. The new energy unit is difficult to provide rotary inertia for the system due to the fact that the new energy unit is incorporated into a power grid through a power electronic converter, so that system inertia response and primary frequency modulation capacity are remarkably reduced, and risks of frequency abnormal fluctuation and instability after system disturbance are increased. Therefore, the rapid frequency prediction after the power system disturbance plays an important role in the safe and stable operation of the system.
The traditional power system frequency prediction method mainly comprises a physical driving method and a data driving method. The physics-driven methods generally describe the characteristics of the study object according to the existing physics knowledge and rules, such as full-time-domain simulation, equivalence modeling, and linearization analysis. The full time domain simulation method has large calculation amount and long time consumption, and is difficult to meet the requirement of quickly predicting the frequency after the system is disturbed. The equivalent model method assumes uniform frequency of the whole network, the calculation precision is low, and the calculation result cannot reflect the time-space distribution characteristic of the system frequency. The computational accuracy of the linearization analysis decreases with increasing degree of simplification. In general, the method for calculating the frequency response of the power system based on the physical driving often has a contradiction between calculation speed and accuracy. Most of data driving methods are 'black box models', namely, the incidence relation between input and output is mined from data without modeling complex physical problems. The data driving method has high dependence on the quantity and quality of data, and when the data quantity is small, the generalization capability of the data driving method is difficult to guarantee, and the interpretability of a calculation result is poor. Currently, many frequency prediction methods based on a physical driving method and a data driving method are studied, but few frequency prediction methods based on a physical-data combined driving method are studied.
Disclosure of Invention
The invention aims to provide a power system frequency prediction method driven by physics-data combination. Performing electromechanical transient simulation on the power system by adopting simulation software to generate power system disturbance data; selecting a first part of input characteristic quantity and output quantity from the generated disturbance data; calculating a second part of input characteristic quantity by using the selected power system frequency-related physical knowledge and the first part of input characteristic quantity; normalizing the input characteristic quantity and the output quantity of the two parts; dividing the normalized data into training samples and testing samples, constructing a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network, and inputting the training samples into the model for training; inputting the test sample into a trained frequency prediction model to realize rapid and accurate prediction of a power system frequency curve; the method specifically comprises the following steps:
step 1: performing electromechanical transient simulation on the power system by adopting simulation software to generate power system disturbance data;
step 2: acquiring a required first part of input characteristic quantity and output quantity from the generated disturbance data;
and step 3: calculating a second part of input characteristic quantity by using the selected power system frequency-related physical knowledge and the first part of input characteristic quantity;
and 4, step 4: normalizing the input characteristic quantity and the output quantity of the two parts;
and 5: dividing the normalized data into training samples and testing samples, constructing a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network, and inputting the training samples into the model for training;
step 6: and inputting the test sample into the trained frequency prediction model to realize the rapid and accurate prediction of the power system frequency curve.
In step 1, the specific operation of generating the disturbance data is as follows: and calling PSS/E simulation software by using Python programming cycle to generate disturbance data in batches, wherein the simulation time set in each simulation is 30s, and the simulation step length is 0.1 s.
In the step 2, the acquired first part of input characteristic quantities comprise a generator inertia constant, generator mechanical power, generator electromagnetic power, a generator damping coefficient and load power; the obtained output quantity is system frequency curve data.
In the step 3, an equation of motion of each generator rotor is selected as the frequency-related physical knowledge of the power system, and the specific formula is as follows:
where H and ω are the generator inertia constant and rotor angular velocity, P, respectivelymAnd PeThe mechanical power and the electromagnetic power of the generator are respectively, and D is a damping coefficient of the generator; and then inputting three characteristic quantity data of the mechanical power of the generator, the electromagnetic power of the generator and the damping coefficient of the generator in the first part input characteristic quantity into a rotor motion equation of the corresponding generator, and calculating to obtain the angular speed of the rotor of each generator, namely the second part input characteristic quantity.
In the step 4, normalization processing is performed on the first and second part input feature quantities and the output quantity by using the following formulas:
in the formula, x' is disturbance data after normalization, x is disturbance data, and x ismaxAnd xminMaximum and minimum values of the disturbance data respectively; normalizing the first and second part of input characteristic quantity and output quantity to [0,1 ] by using Python programming function according to the formula]Within the interval.
In the step 5, a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network is constructed, and a training sample is input into the model for training, and the specific process is as follows: let the training sample input at time t be xtThen h is output at the current momenttThe calculation formula of (2) is as follows:
rt=σ(Wrxxt+Wrhht-1+br)
zt=σ(Wzxxt+Wzhht-1+bz)
where W and b are the weight parameter and bias parameter obtained by training the GRU, respectively. Wrx、WrhAnd brFor calculating the parameter required for resetting the gate output, Wzx、WzhAnd bzFor calculating the parameters required for updating the gate output, Whx、WhhAnd bhParameters required for calculating the process quantities. σ and tanh represent sigmoid and hyperbolic tangent activation functions, respectively. An indication of a matrix corresponding to a position element multiplication. And then inputting the output quantity and the two parts of characteristic quantities into a model for training, setting the number of training iterations to be 1500, and finishing the training process when the number of training iterations reaches 1500.
Step 6, inputting the two parts of characteristic quantities of the test sample into the trained frequency prediction model to obtain a predicted output frequency curve, and comparing the predicted output frequency curve with an output accurate frequency curve of the test sample; in order to comprehensively evaluate the frequency prediction accuracy of the model, an absolute error AE, a root mean square error RMSE, an average absolute error MAE and an average relative error MRE are selected as evaluation indexes; wherein the AE can directly reflect the absolute error between the predicted value and the true value; the RMSE can reflect the discrete degree of a predicted value relative to a real value and can also reflect the centralized degree of a predicted error value; the MAE can reflect the average error between the predicted value and the true value; the MRE can reflect the deviation degree of the predicted value relative to the true value; and then evaluating the prediction curve according to various error evaluation indexes, if the prediction result of the model on the test sample meets the requirements of various error evaluation indexes, training the model to have good prediction performance, if the prediction result of the model on the test sample does not meet the requirements of various error evaluation indexes, repeating the step 5 and the step 6 until the prediction result meets the index requirements, and at the moment, storing the trained model and using the model for frequency prediction.
The invention has the beneficial effects that:
the invention provides a power system frequency prediction method driven by physics-data in a combined mode. The method embeds physical knowledge into a machine learning method, is combined with large running data of a power system, can predict a frequency curve more accurately under the condition of less training data volume, has stronger frequency prediction generalization capability, anti-noise capability and interpretability of a result, and has important significance for safe and stable running of the system.
Drawings
FIG. 1 is a flow chart of a method for predicting a frequency of a power system driven by physical-data combination.
Fig. 2 is a graph comparing the maximum absolute error values of four frequency prediction curves.
FIG. 3 is a comparison of frequency prediction curves for a test sample.
FIG. 4 is a comparison graph of prediction errors of various models with noise. The method comprises the steps of (a), (b), (c) and (d) comparing test results of four prediction methods when noise-containing data are used.
Detailed Description
The invention provides a power system frequency prediction method driven by physics-data combination.
The technical solution in the embodiments of the present invention is clearly and completely described below with reference to the drawings in the embodiments of the present invention,
fig. 1 shows a flow chart of a method for predicting the frequency of a power system driven by physical-data combination, which includes the following steps:
step 1: setting a disturbance scene, and performing electromechanical transient simulation on the power system by adopting simulation software to generate power system disturbance data;
in this step, the specific method for generating the disturbance data is as follows: and calling PSS/E simulation software by using Python programming cycle to generate disturbance data in batches, wherein the simulation time set in each simulation is 30s, and the simulation step length is 0.1 s.
Step 2: acquiring a required first part of input characteristic quantity and output quantity from the generated disturbance data;
in the step, the acquired first part of input characteristic quantities comprise a generator inertia constant, generator mechanical power, generator electromagnetic power, a generator damping coefficient and load power; the obtained output quantity is system frequency curve data.
And step 3: calculating a second part of input characteristic quantity by using the selected power system frequency-related physical knowledge and the first part of input characteristic quantity;
in the step, the motion equation of each generator rotor is selected as the power system frequency-related physical knowledge, and the specific formula is as follows:
where H and ω are the generator inertia constant and rotor angular velocity, P, respectivelymAnd PeThe mechanical power and the electromagnetic power of the generator are respectively, and D is the damping coefficient of the generator. And then inputting three characteristic quantity data of the mechanical power of the generator, the electromagnetic power of the generator and the damping coefficient of the generator in the first part input characteristic quantity into a rotor motion equation of the corresponding generator, and calculating to obtain the angular speed of the rotor of each generator, namely the second part input characteristic quantity.
And 4, step 4: normalizing the input characteristic quantity and the output quantity of the two parts;
in this step, the first and second partial input feature quantities and the output quantity are normalized using the following formulas:
in the formula, x' is disturbance data after normalization, x is disturbance data, and x ismaxAnd xminMaximum and minimum values of the disturbance data, respectively. Writing a function by using Python according to the formulaTwo-part input feature quantity and output quantity are normalized to [0, 1%]Within the interval.
And 5: dividing the normalized data into training samples and testing samples, constructing a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network, and inputting the training samples into the model for training;
in the step, a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network is constructed, and a training sample is input into the model for training, and the specific process is as follows: let the training sample input at time t be xtThen h is output at the current momenttThe calculation formula of (2) is as follows:
rt=σ(Wrxxt+Wrhht-1+br)
zt=σ(Wzxxt+Wzhht-1+bz)
where W and b are the weight parameter and bias parameter obtained by training the GRU, respectively. Wrx、WrhAnd brFor calculating the parameter required for resetting the gate output, Wzx、WzhAnd bzFor calculating the parameters required for updating the gate output, Whx、WhhAnd bhParameters required for calculating the process quantities. σ and tanh represent sigmoid and hyperbolic tangent activation functions, respectively. An indication of a matrix corresponding to a position element multiplication. And then inputting the output quantity and the two parts of characteristic quantities into a model for training, setting the number of training iterations to be 1500, and finishing the training process when the number of training iterations reaches 1500.
Step 6: and inputting the test sample into the trained frequency prediction model to realize the rapid and accurate prediction of the power system frequency curve. The specific process is as follows: and inputting the two parts of characteristic quantities of the test sample into the trained frequency prediction model to obtain a predicted output frequency curve, and comparing the predicted output frequency curve with the output accurate frequency curve of the test sample. In order to comprehensively evaluate the frequency prediction accuracy of the model, an Absolute Error (AE), a Root Mean Square Error (RMSE), an average Absolute Error (MAE), and an average Relative Error (MRE) are selected as evaluation indexes. The AE can directly reflect the absolute error between the predicted value and the true value; the RMSE can reflect the discrete degree of a predicted value relative to a real value and can also reflect the centralized degree of a predicted error value; the MAE can reflect the average error between the predicted value and the true value; the MRE can reflect the degree of deviation of the predicted values from the true values. Taking the single-output predicted value as an example, the definition of the index is as follows:
AE=yi-f(xi)
wherein N is the number of samples, i is the sample number, yiIs the exact value of the ith sample, f (x)i) Is the predicted value of the ith sample. If the prediction result of the model on the test sample meets the requirements of each index, the training model has good prediction performance, if the prediction result does not meet the requirements of each index, the step 5 and the step 6 are repeated until the prediction result meets the requirements of the indexes, and at the moment, the trained model can be stored and used for the frequency prediction problem.
Examples
The example is based on a 39-node system of a new england 10 machine, the system comprises 10 generators, 39 buses, 19 loads and 34 transmission lines, and the reference frequency of the system is 60 Hz. At present, a power system usually contains a high proportion of grid-connected new energy machine sets, and the proportion of the new energy machine sets is gradually increased in the future. In order to consider the influence of a new energy high-permeability scene on the frequency response characteristic of the system, three wind power plants are respectively accessed to nodes No. 2, No. 29 and No. 39, so that the output of a new energy unit is increased from 0% to 30%.
1. The PSS/E software is called circularly through Python programming to obtain the large amount of data needed by the neural network training. The type of disturbance considered herein is load variation, and the specific method is to randomly perform an increasing load disturbance of a single load or a plurality of loads. The perturbation size ranged from 0% to 30%, with 1% increase each, resulting in 2500 sets of samples.
2. Based on the gated cyclic unit neural network method, the first part inputs characteristic quantity as information of disturbance moment and a subsequent period of time, specifically data of the first 6 moments (moment before disturbance, moment after disturbance and four moments after disturbance), and outputs quantity is a system COI frequency curve containing 302 moment data.
3. The second-order rotor motion equations of all generators in the system are selected as physical knowledge, and partial information of the first 6 moments (the first partial input characteristic quantities are the same) in the simulation data is extracted, wherein the partial information specifically comprises mechanical power, electromagnetic power and damping coefficients of the generators. And calculating the angular speeds of all the generator rotors at the first 6 moments, and taking the angular speeds as second part input characteristic quantities. Up to this point, the input feature set including the first and second partial input features has been constructed.
4. In order to evaluate the generalization ability of the model when training samples are small, 20% of the total samples are randomly extracted to construct a small sample set comprising 500 groups of samples, wherein the training samples are 450 groups and the test samples are 50 groups. In order to evaluate the frequency prediction performance of the method, three methods, namely a physical knowledge fusion cyclic neural network (PGRNN), a physical knowledge fusion back propagation neural network (PG-BPNN) and a Back Propagation Neural Network (BPNN), are selected as comparison models, and the prediction performance of the model is more comprehensively evaluated through comparison tests. Fig. 2 shows the maximum absolute error value comparison of the frequency prediction curves, and the following conclusions can be drawn from the four prediction curves in fig. 2: the accuracy of the predicted frequency curve of the method is higher, which shows that the method can obtain better accuracy under the condition of less training samples; embedding physical knowledge helps to improve prediction accuracy.
5. In order to observe the prediction accuracy of the frequency response curve more clearly, test sample No. 13 was selected as an evaluation object, and the prediction curves of the four prediction methods were compared, as shown in fig. 3. The result shows that the difference between the frequency curve predicted by the method and the accurate value is minimum, so the prediction accuracy is highest. In addition, table 1 summarizes the evaluation results of the three error indicators RMSE, MAE, MRE of the prediction curves. The result shows that compared with the other three methods, the method has lower errors under the three evaluation standards, and further proves that the method has better frequency prediction performance.
TABLE 1 comparison of three evaluation indices of frequency prediction curves
6. Model anti-noise capability assessment
To evaluate the noise immunity of the model, noise conforming to a gaussian distribution was added to the data to better simulate the actual situation. The specific method is as follows:
data_noise=data(1+θ)
in the formula, data _ noise represents data after noise is added, data represents original measurement data, and theta represents Gaussian noise with a mean value of 0 and a variance of beta. FIG. 4 shows the comparison of the test results of the four prediction methods (a), (b), (c) and (d) when using the noise-containing data. The results show that: when the noise is gradually increased, the prediction error is not obviously increased, so that the noise resistance of the method is strong.
Claims (7)
1. A power system frequency prediction method driven by physics-data combination. The method is characterized by comprising the following steps:
step 1: performing electromechanical transient simulation on the power system by adopting simulation software to generate power system disturbance data;
step 2: acquiring a required first part of input characteristic quantity and output quantity from the generated disturbance data;
and step 3: calculating a second part of input characteristic quantity by using the selected power system frequency-related physical knowledge and the first part of input characteristic quantity;
and 4, step 4: normalizing the input characteristic quantity and the output quantity of the two parts;
and 5: dividing the normalized data into training samples and testing samples, constructing a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network, and inputting the training samples into the model for training;
step 6: and inputting the test sample into the trained frequency prediction model to realize the rapid and accurate prediction of the power system frequency curve.
2. The method for predicting the frequency of the power system driven by the physical-data combination according to claim 1. The method is characterized in that in the step 1, the specific operation of generating the disturbance data is as follows: and calling PSS/E simulation software by using Python programming cycle to generate disturbance data in batches, wherein the simulation time set in each simulation is 30s, and the simulation step length is 0.1 s.
3. The method for predicting the frequency of the power system driven by the physical-data combination according to claim 1. The method is characterized in that in the step 2, the acquired first part of input characteristic quantities comprise a generator inertia constant, generator mechanical power, generator electromagnetic power, a generator damping coefficient and load power; the obtained output quantity is system frequency curve data.
4. The method for predicting the frequency of the power system driven by the physical-data combination according to claim 1. The method is characterized in that in the step 3, a motion equation of each generator rotor is selected as the power system frequency-related physical knowledge, and the specific formula is as follows:
where H and ω are the generator inertia constant and rotor angular velocity, P, respectivelymAnd PeThe mechanical power and the electromagnetic power of the generator are respectively, and D is a damping coefficient of the generator; and then inputting three characteristic quantity data of the mechanical power of the generator, the electromagnetic power of the generator and the damping coefficient of the generator in the first part input characteristic quantity into a rotor motion equation of the corresponding generator, and calculating to obtain the angular speed of the rotor of each generator, namely the second part input characteristic quantity.
5. The method for predicting the frequency of the power system driven by the physical-data combination according to claim 1. In step 4, normalization processing is performed on the first and second partial input feature quantities and the output quantity by using the following formulas:
in the formula, x' is disturbance data after normalization, x is disturbance data, and x ismaxAnd xminMaximum and minimum values of the disturbance data respectively; normalizing the first and second part of input characteristic quantity and output quantity to [0,1 ] by using Python programming function according to the formula]Within the interval.
6. The method for predicting the frequency of the power system driven by the physical-data combination according to claim 1. The method is characterized in that in the step 5, a power system frequency prediction model based on a physical-data drive gate control cycle unit neural network is constructed, and a training sample is input into the model for training, and the specific process is as follows: assume training sample input at time tIs xtThen h is output at the current momenttThe calculation formula of (2) is as follows:
rt=σ(Wrxxt+Wrhht-1+br)
zt=σ(Wzxxt+Wzhht-1+bz)
wherein, W and b are weight parameters and bias parameters obtained by training GRUs respectively; wrx、WrhAnd brFor calculating the parameter required for resetting the gate output, Wzx、WzhAnd bzFor calculating the parameters required for updating the gate output, Whx、WhhAnd bhParameters required for calculating process quantities; sigma and tanh represent sigmoid and hyperbolic tangent activation functions, respectively; an indication of a multiplication of corresponding position elements of the matrix; and then inputting the output quantity and the two parts of characteristic quantities into a model for training, setting the number of training iterations to be 1500, and finishing the training process when the number of training iterations reaches 1500.
7. The method for predicting the frequency of the power system driven by the physical-data combination according to claim 1. The method is characterized in that in the step 6, the two parts of characteristic quantities of the test sample are input into a trained frequency prediction model to obtain a predicted output frequency curve, and the predicted output frequency curve is compared with an accurate output frequency curve of the test sample; in order to comprehensively evaluate the frequency prediction accuracy of the model, an absolute error AE, a root mean square error RMSE, an average absolute error MAE and an average relative error MRE are selected as evaluation indexes; wherein the AE can directly reflect the absolute error between the predicted value and the true value; the RMSE can reflect the discrete degree of a predicted value relative to a real value and can also reflect the centralized degree of a predicted error value; the MAE can reflect the average error between the predicted value and the true value; the MRE can reflect the deviation degree of the predicted value relative to the true value; and then evaluating the prediction curve according to various error evaluation indexes, if the prediction result of the model on the test sample meets the requirements of various error evaluation indexes, training the model to have good prediction performance, if the prediction result of the model on the test sample does not meet the requirements of various error evaluation indexes, repeating the step 5 and the step 6 until the prediction result meets the index requirements, and at the moment, storing the trained model and using the model for frequency prediction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110922955.2A CN113837432B (en) | 2021-08-12 | 2021-08-12 | Power system frequency prediction method driven by physical-data combination |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110922955.2A CN113837432B (en) | 2021-08-12 | 2021-08-12 | Power system frequency prediction method driven by physical-data combination |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113837432A true CN113837432A (en) | 2021-12-24 |
CN113837432B CN113837432B (en) | 2023-04-25 |
Family
ID=78963251
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110922955.2A Active CN113837432B (en) | 2021-08-12 | 2021-08-12 | Power system frequency prediction method driven by physical-data combination |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113837432B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114583767A (en) * | 2022-03-10 | 2022-06-03 | 中国电力科学研究院有限公司 | Data-driven wind power plant frequency modulation response characteristic modeling method and system |
CN117937521A (en) * | 2024-03-25 | 2024-04-26 | 山东大学 | Power system transient frequency stability prediction method, system, medium and equipment |
WO2024092322A1 (en) * | 2022-11-03 | 2024-05-10 | Firmus Technologies Pty Ltd | "methods and systems for providing stability to an electricity grid" |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104333005A (en) * | 2014-08-18 | 2015-02-04 | 西南交通大学 | Electrical-power-system post-disturbance frequency dynamic-state prediction method based on support vector regression |
US20190286971A1 (en) * | 2018-03-15 | 2019-09-19 | Advanced Micro Devices, Inc. | Reconfigurable prediction engine for general processor counting |
CN112883522A (en) * | 2021-01-14 | 2021-06-01 | 沈阳工业大学 | Micro-grid dynamic equivalent modeling method based on GRU (generalized regression Unit) recurrent neural network |
-
2021
- 2021-08-12 CN CN202110922955.2A patent/CN113837432B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104333005A (en) * | 2014-08-18 | 2015-02-04 | 西南交通大学 | Electrical-power-system post-disturbance frequency dynamic-state prediction method based on support vector regression |
US20190286971A1 (en) * | 2018-03-15 | 2019-09-19 | Advanced Micro Devices, Inc. | Reconfigurable prediction engine for general processor counting |
CN112883522A (en) * | 2021-01-14 | 2021-06-01 | 沈阳工业大学 | Micro-grid dynamic equivalent modeling method based on GRU (generalized regression Unit) recurrent neural network |
Non-Patent Citations (1)
Title |
---|
王琦 等: "基于物理—数据融合模型的电网暂态频率特征在线预测方法", 《电力系统自动化》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114583767A (en) * | 2022-03-10 | 2022-06-03 | 中国电力科学研究院有限公司 | Data-driven wind power plant frequency modulation response characteristic modeling method and system |
CN114583767B (en) * | 2022-03-10 | 2023-03-17 | 中国电力科学研究院有限公司 | Data-driven wind power plant frequency modulation response characteristic modeling method and system |
WO2024092322A1 (en) * | 2022-11-03 | 2024-05-10 | Firmus Technologies Pty Ltd | "methods and systems for providing stability to an electricity grid" |
CN117937521A (en) * | 2024-03-25 | 2024-04-26 | 山东大学 | Power system transient frequency stability prediction method, system, medium and equipment |
Also Published As
Publication number | Publication date |
---|---|
CN113837432B (en) | 2023-04-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113837432B (en) | Power system frequency prediction method driven by physical-data combination | |
CN109117951B (en) | BP neural network-based probability load flow online calculation method | |
CN111179121B (en) | Power grid emergency control method based on expert system and deep reverse reinforcement learning | |
CN107563067A (en) | Analysis of structural reliability method based on Adaptive proxy model | |
CN110119570B (en) | Actually measured data driven wind farm model parameter checking method | |
CN111426957B (en) | SOC estimation optimization method for power battery under simulated vehicle working condition | |
CN111415010B (en) | Bayesian neural network-based wind turbine generator parameter identification method | |
CN109063939A (en) | A kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network | |
CN116707331B (en) | Inverter output voltage high-precision adjusting method and system based on model prediction | |
CN114006370B (en) | Power system transient stability analysis and evaluation method and system | |
CN112883522A (en) | Micro-grid dynamic equivalent modeling method based on GRU (generalized regression Unit) recurrent neural network | |
CN110634082A (en) | Low-frequency load shedding system operation stage prediction method based on deep learning | |
CN110765703A (en) | Wind power plant aggregation characteristic modeling method | |
CN114583767B (en) | Data-driven wind power plant frequency modulation response characteristic modeling method and system | |
CN117077546B (en) | Power system load modeling method and system based on data driving | |
Zhang et al. | A method based on Lorenz disturbance and variational mode decomposition for wind speed prediction | |
Lin et al. | Uncertainty quantification in dynamic simulations of large-scale power system models using the high-order probabilistic collocation method on sparse grids | |
Xu et al. | Short-term wind speed prediction based on GRU | |
CN115659844B (en) | Simulation method and device based on wind power plant dynamic model and computer equipment | |
Zhang et al. | Prediction for the maximum frequency deviation of post-disturbance based on the deep belief network | |
CN116226620A (en) | Subsynchronous oscillation mode identification method based on SVM and LSTM | |
CN105930613A (en) | Equivalent modeling method for distributed power generation system | |
Chen et al. | A Method of Frequency Security Margin Index Prediction of Post-disturbance Power System Based on Classification and Regression Model of XGBoost Algorithm | |
CN113255213B (en) | Power converter composite fault identification method | |
Ding et al. | Data-Driven Dynamic Modeling Methods for Offshore Wind Turbine Generators |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |