CN115187154B - Neural network-based regional power grid oscillation source risk prediction method and system - Google Patents
Neural network-based regional power grid oscillation source risk prediction method and system Download PDFInfo
- Publication number
- CN115187154B CN115187154B CN202211117058.5A CN202211117058A CN115187154B CN 115187154 B CN115187154 B CN 115187154B CN 202211117058 A CN202211117058 A CN 202211117058A CN 115187154 B CN115187154 B CN 115187154B
- Authority
- CN
- China
- Prior art keywords
- node
- matrix
- power grid
- oscillation source
- oscillation
- 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.)
- Active
Links
- 230000010355 oscillation Effects 0.000 title claims abstract description 151
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 20
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 24
- 238000009826 distribution Methods 0.000 claims abstract description 22
- 230000001932 seasonal effect Effects 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000012502 risk assessment Methods 0.000 claims abstract description 7
- 238000013461 design Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 89
- 230000006870 function Effects 0.000 claims description 38
- 230000004913 activation Effects 0.000 claims description 14
- 230000015654 memory Effects 0.000 claims description 13
- 230000003044 adaptive effect Effects 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 230000006835 compression Effects 0.000 claims description 7
- 238000007906 compression Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 7
- 238000013016 damping Methods 0.000 claims description 6
- 238000000354 decomposition reaction Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 230000007613 environmental effect Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims 4
- 125000004122 cyclic group Chemical group 0.000 claims 1
- 239000000126 substance Substances 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 4
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000012545 processing Methods 0.000 description 17
- 238000011161 development Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 238000001556 precipitation Methods 0.000 description 3
- 230000005855 radiation Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000001360 synchronised effect Effects 0.000 description 3
- 230000002123 temporal effect Effects 0.000 description 3
- 238000007792 addition Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000007123 defense Effects 0.000 description 2
- 238000009776 industrial production Methods 0.000 description 2
- 230000002265 prevention Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000009021 linear effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000003071 parasitic effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000009466 transformation 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
- G06Q10/06—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
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Development Economics (AREA)
- Molecular Biology (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Educational Administration (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a neural network-based regional power grid oscillation source risk prediction method and system. Firstly, drawing an approximate 'electrical node virtual plane map', then generating a self-adaptive power grid planarization disturbance image of a convolutional neural network, and finally constructing an oscillation source risk analysis space-time distribution prediction model based on a ConvLSTM network for prediction based on ConvLSTM network design. The invention has the following advantages: 1. the three-dimensional power grid structure can be approximately depicted into an acceptable plane graph, and the data coding form is simplified. 2. The method can effectively compress the graphic data by combining the structural characteristics of the specific power grid and seasonal differences, and improves the training efficiency. 3. The speed of the oscillation source identification and prediction can be greatly improved. 4. The model predicts the influence of the risk quantity of the regional power grid oscillation source, and the prediction effect is better in the time period with larger risk quantity of the regional power grid oscillation source.
Description
Technical Field
The invention relates to a risk prediction and control method for an oscillation source of a regional power grid, in particular to a risk prediction method and a risk prediction system for the oscillation source of the regional power grid based on a neural network.
Background
With the continuous development of computer technology, the information amount of each industry is increased sharply, information which is large in scale and cannot be managed and processed in a reasonable time through a conventional software tool is generally called big data, and currently, the characteristics of the big data are generally considered in the industry to be classified into 4 pieces of V' -Volume (large amount), velocity (high speed), variety (diverse), and Value (low Value). Sensor data is one of main sources of big data, and is a development trend and a key research direction of information technology for big data analysis and application of a sensor network, and the basic technology of the sensor network in a big data environment is deeply researched by LiuL and the like.
The large-scale power system stable defense system needs to continuously acquire and analyze data of sensors related to PMU sensors, has the characteristic of big data naturally, and has the following characteristics besides the traditional big data characteristic (1) that the volume is huge and the data produced by empty management for one year is above PB level; (2) the source is real, PMU high-precision sampling data of each transformer substation and each power plant of the power grid are directly collected to the provincial regional power grid central control system, and the source is real and reliable.
Modern power grids are essentially electric energy systems that are "forced" to operate at 50/60Hz (alternating current) and 0Hz (direct current), but when discussing their oscillation problems, it is usually referred to as "parasitic" or mechanical, or electromagnetic or their coupled reciprocal energy exchanges outside the operating frequency, which cause stability or power quality problems when they jeopardize the normal operation of the power system. Since the birth of the power system, the oscillation is one of the important sides of the dynamic or stability research.
After long-term research, the mechanism and the characteristics of low-frequency oscillation and subsynchronous resonance/oscillation are fully disclosed, and the common characteristics of the low-frequency oscillation and the subsynchronous resonance/oscillation are as follows: the dominance and participation of rotating units with large physical inertia, especially large synchronous generator sets. But recently, the power system is undergoing a deep revolution, and one of the outstanding features and development trends is the wide access of the power electronic converter; on the power supply side, the variable-current power supply is continuously increased, and the ratio of wind power to photovoltaic in a newly-increased installation machine in China exceeds that of a coal-fired unit in 2016 years, and reaches 41.8%; on the power grid side, extra-high voltage direct current, flexible direct current and flexible alternating current power transmission equipment based on a converter is widely applied; and on the user side, the distributed power generation, direct-current distribution network and micro-grid technology of the converter are adopted for vigorous development. These, the dynamic behavior of the power system is changing significantly, bringing new stability and oscillation problems. In recent years, the problems of novel subsynchronous/supersynchronous oscillation caused by a variable-current power supply such as wind power and the like are very outstanding, the negative resistance characteristic of a variable-current constant-power load, the phase-locked loop coupling of a multi-converter, the serial/parallel resonance of the converter control participating in the power grid side, and the interaction between a static synchronous compensator (STATCOM) and a high-voltage direct current transmission (VSC-HVDC) based on a voltage source converter and a weak alternating current power grid once excite the broadband oscillation with the frequency from several Hz to over thousand Hz, and in addition, the problems of harmonic amplification or strong oscillation and the like caused by the converter participation in a power distribution system are caused, thereby bringing about wide attention in the academic and industrial fields.
Regional oscillations induced by a continuous oscillation source have received increasing attention in recent years.
How to evaluate, predict and defend the risk of the oscillation source of the regional power grid becomes an important problem for the safety and stability control of the power system under the condition of a novel power system. This problem can be mainly resolved into: (1) Positioning a historical oscillation source, (2) probabilistic early warning of the oscillation source, and (2) prevention control and defense of risk in advance.
In view of PMU radiation of the existing power grid and the processing performance of a WAMS control system, partial regional power grids have the condition of recording the whole-grid oscillation in the process of generating the oscillation source of the power grid in a panoramic mode, so that data-driven processing is carried out on the oscillation of the oscillation source type power grid from the view point of image processing, and the oscillation source is positioned, and the generation probability is evaluated and prevented and controlled.
ConvLSTM was originally proposed by the article "A Machine Learning Approach for Precipitation Nowcasting" and is intended to solve the problem of Precipitation forecasting. The precipitation forecast problem is generally regarded as a temporal problem and is therefore considered to be solved using LSTM, but pure LSTM cannot utilize spatial data features through pictures, so spatial features are not fully utilized in this LSTM approach. According to the above description, the thesis proposes a ConvLSTM structure, which not only can establish a temporal relationship similar to LSTM, but also can possess a spatial feature extraction capability similar to CNN. And the authors experimentally demonstrated that ConvLSTM has a better effect than LSTM in obtaining spatio-temporal relationships. Moreover, convLSTM can not only predict weather, but also solve the prediction problem of other space-time sequences, such as video classification, action recognition and the like.
In order to improve the safe and stable operation level of the system and perform risk early warning and operation guidance on the regional power grid, it is necessary to predict and control the risk of the regional power grid oscillation source of the system. Along with the replacement of novel power electronic equipment such as new energy power generation, high-voltage direct-current transmission and the like for the power generation of the traditional synchronous generator set, the system broadband oscillation is aggravated, and according to observable data, the small disturbance appears to present an exponential growth trend. The number and the coverage of regional power grid disturbance are larger and larger due to the reasons, the oscillation source is difficult to capture, and the prevention and the control of the local power grid disturbance are difficult.
Under the background, the traditional system for predicting the risk of the regional power grid oscillation source has great error and cannot meet the operation requirement of the system, and a system for predicting and controlling the risk of the regional power grid oscillation source is not considered at home and abroad at present.
Disclosure of Invention
The technical problem of the invention is mainly solved by the following technical scheme:
a method for predicting and controlling risks of regional power grid oscillation sources based on a convolution long-short term memory neural network comprises the following steps:
drawing an electrical node virtual plane map based on the node relation matrix, specifically, equating a three-dimensional power grid network node structure to an equivalent plane power grid network node structure by circularly and equivalently modifying the node relation matrix;
acquiring oscillation source data of an equivalent plane power grid node structure, preprocessing the oscillation source data, outputting the preprocessed oscillation source data to a seasonal adaptive CNN network model, and training the preprocessed oscillation source data to obtain oscillation source classification data suitable for different seasonal periods of a set power grid;
an oscillation source risk analysis space-time distribution prediction model based on a ConvLSTM network is built, classification data of the oscillation source are divided into a training set and a testing set to train and test the prediction model, an optimized prediction model is obtained, and the prediction model can predict and obtain an oscillation disturbance quantity form graph which is possible to occur in the second day after a given daily oscillation record number is input.
In the method for predicting and controlling risk of the oscillation source of the regional power Grid based on the convolutional long-short term memory neural network, the Grid node association admittance matrix of the power Grid is a-Grid node number N:
In the above method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolution long-short term memory neural network,
the row and column ranges are specifically configured as follows:
all 500kV nodes are arranged in a line from 1-N _500 to 200 kV electrical nodes connected with 220kV and 500kV from N _500+1 to N_220 _500, from N _220_500+1 to N _220 _220are common nodes of 220kV, and from N _220+1 to N are common nodes of 110 kV; the diagonal element defaults to 1; other principles for arbitrary
If a direct electrical line link exists between the node i and the node j, the value is set to 1, otherwise, the value is set to 0.
In the above method for predicting and controlling the risk of the regional power grid oscillation source based on the convolution long and short term memory neural network, during equivalent transformation,
determining a plurality of 500 kilovolt electrical nodes to which any one 110 kilovolt electrical node is subordinate to form a matrix T _5_1, namely a 3 x N \500 matrix;
traversing a given 110KV electrical node, inquiring non-0 elements of a corresponding row of a node i in an initial node relation matrix, and deleting the row except for a diagonal array;
determining pairs of adjacent 500kV electrical nodes by using a matrix T _5_1, and determining matrix element values corresponding to the associated electrical node nodes by using the following formula until all 110kV electrical nodes are traversed;
wherein, the first and the second end of the pipe are connected with each other,
。
in the method for predicting and controlling the risk of the regional power grid oscillation source based on the convolution long-short term memory neural network, the CNN network model comprises an oscillation preprocessing layer, and the CNN network model is used for preprocessing oscillation source data, namely, image compression is carried out after daily disturbance pattern scanning is carried out by adopting an activation function, specifically, the CNN network model is used for compressing images after daily disturbance pattern scanning
Creating a 2D geographic matrix with an image matrix of M x M, wherein the 2D geographic matrix is in single mapping with a node relation matrix corresponding to nodes;
changing the image matrix A _ Grid into A _ oc _ Grid by adopting an improved sigmoid activation function:
wherein, according to the classical theory of forced power oscillation,
wherein A is the maximum amplitude after PRONY decomposition, and
is the damping value corresponding to the maximum amplitude after decomposition.
Compressing a blank invalid region of the image in the image matrix A _ oc _ Grid, namely deleting all blank rows and columns in the image matrix A _ oc _ Grid to generate A _ oc _ Grid _ dense.
In the method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolution long-short term memory neural network, the CNN network model further comprises at least two CONV layers, the input is regional terrain elements, the output is processed regional terrain elements, and the activation function of each layer is an improved relu function:
n is a positive integer and is determined by combining artificial experience and retrograde motion according to different seasons.
In the method for predicting and controlling the risk of the regional power grid oscillation source based on the convolution long-short term memory neural network, an attention mechanism layer is added behind a full connection layer in a prediction model to serve as an output, namely a Softmax function layer is added behind the full connection layer, and the Softmax function layer is defined as follows:
wherein
And C is the number of output nodes, namely the number of output of the full connection layer, namely the number of classified categories.
In the method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolution long-short term memory neural network, input data of a prediction model is a 3D image matrix of L x L, and corresponding nodes in the 3D matrix and corresponding nodes in a 2D geographic matrix are in single mapping.
In the method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolution long-short term memory neural network, one element is arranged in any t dimension in a 3D geographic matrix with a 3D matrix C structure of L x D
The numerical value is represented by the number of the single-day oscillation amplitude values recorded by the node, t is the time dimension, the time scale is 1 day, the predicted time length is 7 days, and t =7.
A system, comprising
A first module: the node relation matrix is configured to draw an electrical node virtual plane map based on the node relation matrix, and specifically, the node relation matrix is modified circularly and equivalently to enable a three-dimensional power grid network node structure to be equivalent to an equivalent plane power grid network node structure;
a second module: the system comprises a network node structure, a seasonal adaptive CNN network model and a power grid, wherein the network node structure is configured to acquire oscillation source data of an equivalent plane power grid network node structure, preprocess the oscillation source data and output the oscillation source data to train the oscillation source data in the seasonal adaptive CNN network model, so that oscillation source classification data suitable for setting different seasonal periods of the power grid are acquired;
a third module: the method is configured to construct a ConvLSTM network-based oscillation source risk analysis space-time distribution prediction model, the prediction model is trained and tested by dividing oscillation source classification data into a training set and a testing set, the optimized prediction model is obtained, and the prediction model can predict and obtain an oscillation disturbance quantity form graph which is possible to occur in the next day after a given daily oscillation record number is input.
Therefore, the invention has the following advantages: 1. the three-dimensional power grid structure can be approximately depicted into an acceptable plane graph, the relevance between the oscillation identification and the regional oscillation source evaluation is established, the data coding form is simplified, and the power grid oscillation source distribution and the diffusion distribution thereof can be depicted in a simplified mode. 2. The intelligent training system can combine the structural characteristics of a specific power grid and seasonal differences to effectively compress graphic data, and improves training efficiency on the premise of ensuring precision. 3. By adopting a data driving mode, the speed of identification and prediction of the oscillation source can be greatly improved within an acceptable range of regional power grid risk management and control in precision. 4. The model predicts the influence of the risk quantity of the regional power grid oscillation source, and the prediction effect is better in the time period with larger risk quantity of the regional power grid oscillation source. 5. Meanwhile, the ConvLSTM network-based regional power grid oscillation source risk space-time distribution prediction model and other deep learning-based prediction models are compared and analyzed, and the result shows that the prediction model built in the ConvLSTM network model is superior to the CNN prediction model based on the regional power grid oscillation source risk and the FC-LSTM prediction model based on the regional power grid oscillation source risk.
Drawings
FIG. 1 is a schematic flow chart of the present invention for preprocessing data.
FIG. 2 is a schematic diagram of the process of image compression after scanning the daily disturbance pattern in the present invention.
Detailed Description
The technical scheme of the invention is further specifically described by the following embodiments and the accompanying drawings.
Example (b):
first of all, the first step is to,
Because a relatively mature (in-line) convolutional neural network mainly processes a plane image, and an actual electric network is a multi-voltage-class (including 500kV alternating current, 220kV alternating current and 110kV alternating current) electromagnetic ring network structure, namely a three-dimensional power grid network structure, the convolutional neural network cannot be directly used in the problem of oscillation source analysis, so that the problem of power grid network disturbance identification and other related problems by adopting an image processing artificial intelligence method become bottlenecks. However, for provincial regional power grids in China, a large number of open-loop structures exist, and most regional power grid structures have 2 characteristics: 1) (ii) a 110 kilovolt and below are all radial power grids, and 220 kilovolt and above are electromagnetic ring networks; 2) (ii) a The provincial region power grid usually takes a 500kV power grid as a backbone network frame and is divided into a plurality of 500 kV/220 kV electromagnetic ring network sections, the electromagnetic ring network sections are only connected with one another through 500kV voltage grade lines, 220kV inside each electromagnetic ring network section is in a ring network structure, and meanwhile, the 110kV power grid is in a radial structure, so that the step mainly comprises the following sub-steps.
Step 1.1, defining a node relation matrix.
For the oscillation tracing of the regional power grid, the method has the characteristics of industrial production: that is, in the field of power system industrial production, because the new energy source accessed to the system or the electrical topology position of the thermal power generating unit is priori knowledge, the accuracy of the positioning of the oscillation source does not need to be positioned to a specific electrical node, and when the oscillation source can be locked to a given countable possible oscillation source occurrence area, the occurrence place of the oscillation source can be basically judged and a corresponding control or suppression strategy can be implemented.
Therefore, the following simplified processing is carried out on the given power grid network, the three-dimensional electric network is converted and approximately converted into the planar electric diagram on the premise of ensuring that the oscillation source can be locked to a certain potential oscillation source area with a priori knowledge, and the subsequent processing speed of the oscillation source is improved.
Let the node association (admittance matrix) matrix of Grid of an actual given electrical network be a _ Grid and the number of nodes be N. Then the formula is shown as (1-1).
For arbitrary
Representing the electrical association relationship between the node i and the node j, considering the basic rule of the electric network, in the invention, all 500kV nodes are arranged in a line from 1-N _500 from N _500+1 to N _220 _500to 200 kV electrical nodes formed by connecting 220kV and 500kV, from N _220_500+1 to N _220 _220to 220kV common nodes, and from N _220+1 to N to 110kV common nodes; the diagonal element defaults to 1; other principles for arbitrary
And when a direct electric line link exists between the node i and the node j, setting the value to be 1, otherwise, setting the value to be 0.
The aforementioned N _500, N _220_500, N _220, N are determined depending on the number of specific network nodes.
And 1.2, circularly and equivalently modifying the node relation matrix. For the relation of the provincial regional power grid electrical nodes represented by the formula (1-1), the three-dimensional structure of the three-dimensional structure mainly exists in each 500 kV/220 kV electromagnetic ring network section, and for common 220 common electrical nodes of the 500 kV/220 kV electromagnetic ring network section
The planarization process therefore consists primarily in establishing a home relationship between 110kv node pairs and 500kv nodes. The formula (1-1) before modification is characterized in that: two parts of common nodes are ring-shaped structures from N _500+1 to N _220 _500to 200 KV electrical nodes connected with 220KV and 500KV, and from N _220_500+1 to N _220to 220KV common nodes, so that the addition of row or column elements in any principle is more than or equal to 3; and for the nodes of 110kv part, the sum of the row or column principle is necessarily 2 because the layer is in a radial grid structure.
As all oscillation sources of the actual power grid are thermal power plants, new energy stations and direct current feed-in stations generally and serve as universal alternating current power grids, node voltage modulus phase angles
The structure diagram of the circulation processing of the method of the invention is shown in figure 1. The criterion is expressed as formula (1-2), which is called as 'disturbance associated activation function' in the invention,
since the matrix is a symmetric matrix, therefore
。
Through the above steps, the original node relation matrix is converted into a matrix which is between 0 and 2.5 for the 110kV node element row and column and is only associated with 500kV elements in the non-0 principle. Meanwhile, the 220kv element relates to the matrix part, and the other row and column elements are 0 except the diagonal element is 1.
From the network topology point of view, the regional power grid topology is degraded to a topology with only 500kv nodes and 110kv nodes, wherein 500kv layer is ring network structure, 110kv nodes are radial, and radiation convergence points are all 500kv nodes.
And step 1.3, matrix compression.
Through the modification in the step 1.2, the original node relation matrix is converted into an extremely sparse matrix, the matrix information of the 220KV part is invalid information actually, and the node relation matrix information can be greatly simplified through deletion operation on the premise of ensuring the acceptable degree of the oscillation source identification, so that the efficiency of the subsequent image processing operation steps is improved.
And 2, generating a planar disturbance image of the convolutional neural network adaptive power grid (elliptic mapping mirror surface).
The power grid oscillation generating source and the radiation range thereof have a fixed regular rule and a seasonal rule, different regional power grids have different geographical ranges due to different provinces and light resources, and the imbalance of the traditional thermal power generating unit and load distribution, the oscillation sources concerned by different regional power grids have different emphasis points, and if the same resolution ratio is adopted for analysis at the same time, the calculation resources are greatly consumed.
In the step, historical big data are established by relying on the improved node relation matrix established in the step 1, and the historical big data are input into an oscillation area specially used for identifying key points. Therefore, historical oscillation source data is input into a convolutional neural network, key areas are identified, time characteristics are marked, and different oscillation source geographic maps are adopted in a given time period. This step mainly includes the following substeps.
Step 2.1, image compression based on the oscillation history quantity (adding an oscillation pattern deformation layer before injecting into the input layer of the convolutional neural network, or an image mode).
The daily perturbation pattern is given an activation function sweep using the activation function.
Step 2.1.1, create 2D geographical matrix with image matrix M x M. And (3) forming a single mapping relation between the actual corresponding node and the node relation matrix in the step (1). For any one element in the matrix of the M x M2D geographic matrix
And the value is represented as the amplitude value of the maximum amplitude value obtained by PRONY analysis of an oscillation scene recorded by the node.
Step 2.1.2, create a dynamic adaptive activation function (oscillation pre-processing layer).
And taking a grid acceptable threshold value between the maximum value and the minimum value as 0.
Improving the sigmoid activation function:
wherein, according to the classical theory of forced power oscillation,
whereinAMaximum amplitude after PRONY decomposition, and
the maximum amplitude damping value after decomposition is used, so that the function is activated, the larger the amplitude of a given image element is after the improved sigmoid function, the larger the damping value with weaker damping can be obtained, and the numerical value is larger (relative to the oscillation disturbance record with stronger damping), and the image processing facing the power grid oscillation source identification is facilitated.
And changing the image matrix processed by the improved sigmoid function from A _ Grid to A _ oc _ Grid.
And 2.1.3, compressing the image with the highlighted graphic emphasis based on the processing of 2.1.2.
In view of different networks and different seasons of the networks, the oscillation source and the development pattern thereof are often particularly in the key area, and when the fixed convolution processing is carried out, if the non-key area is scanned, a large number of managers are consumed. The relevant regions can be locked in for detailed analysis using a priori knowledge or sample image analysis.
All the collected A _ oc _ Grid are used as Sample sets, and the number of the Sample sets is set as N _ Sample.
In the Sample picture of N _ Sample, the image itself cannot be compressed any more in consideration of differences in the number rotation and the trajectory distortion, compared to the normal image recognition (for example, character recognition). However, the geographical wiring diagram of the WAMS oscillation system has relatively fixed electrical geographical nodes, and does not rotate or artificially distort, and the difference mainly lies in the difference of oscillation generation sources and the oscillation size. Therefore, the blank (invalid area) of the image can be further compressed, and the compression encoding flow is as shown in fig. 2.
And 2.2, carrying out recognition, classification, training and determination on the disturbance source based on the seasonal adaptive traditional CNN network.
The initial basic parameters are: inputting the regional terrain elements into the CONV layer, wherein the CONV can be divided into two layers at least through the efficiency improving processing, the initial filters =16 adopted by the first layer, the convolution kernel size is 3 x 3, the activation function is an improved relu function, the filters =32 adopted by the second layer, the convolution kernel size is 7 x 7, and the activation function is an improved relu function; and outputting the processed regional terrain elements (namely terrain influence elements) through the two CONV layers. The above parameters may be adjusted according to the specific map shape.
And (4) obtaining disturbance source geographical wiring diagrams of different distribution area identification degree combinations through the processing of the steps.
According to different seasons, the oscillation forms in different seasons are different greatly due to the starting mode, load change and the like, and the oscillation forms in the same season have a certain fixed regular pattern.
The classical relu function is defined as follows:
the relu function is the most commonly used activation function in the field of current convolutional neural networks, and its mathematical description is as follows:
the output value of the function is in the range of 0 to infinity, and the function is essentially a comparative piecewise function when
When the function value is 0 and the gradient is 0, and when
When the function output value is
And the gradient is 1. The relu function perfectly circumvents the gradient vanishing phenomenon. The disadvantages are that: but when the input value of the neuron is less than 0, the corresponding weight will not be updated again, and unfortunately, the relu functionThe output is not 0 as the center, namely the distribution of the input data is changed, the distribution of the input data of the next layer is different from the distribution of the input data of the previous layer, the training speed of the model can be greatly reduced, and the training speed is slowed down because the model needs to be continuously adapted to different input distributions. Although the relu function has the above disadvantages in the general case, the image processing data is greatly simplified by combining the target problem characteristics through the processing from step 1 to step 2.1 when the image processing established by the present invention is processed, so that the training speeds of the image processing data are possibly mutually offset.
The definition of the modified relu function is described as follows:
first, the expressions (1 to 4) can be converted into the expressions of the following expressions (1 to 5).
Considering the inventive variant as shown in equations (1-6):
when x is close to the range of 0-1, the function of x and the traditional relu function show similar effect, namely linear effect; and once the output is more than 1, the output of the oscillator shows an exponential type ascending trend so as to better highlight the oscillating characteristic. n is a positive integer and is determined by combining artificial experience and retrograde motion according to different seasons.
Through the training of the steps, the classification of the oscillation source suitable for different seasonal periods of a specific power grid can be trained.
And 3, designing and constructing a ConvLSTM network and an oscillation source risk analysis space-time distribution prediction model based on the ConvLSTM network.
Through the four parts of ConvLSTM network design, relevant environmental impact factor fusion and prediction result evaluation methods, a regional power grid oscillation source space-time distribution prediction model based on the ConvLSTM network is constructed, and the problem that sparse regional power grid oscillation source traceability and divergence distribution characteristics are difficult to identify is solved.
And 3.1, creating a 3D matrix with an image matrix of L.
And (3) forming a single mapping relation between the actual corresponding node and the matrix b of the node relation matrix in the step (2).
Setting any t dimension in the 3D geographic matrix with the structure of the 3D matrix C as L D
The numerical value is expressed as the number of single-day oscillation amplitude values recorded by the node, t is a time dimension, the time scale is 1 day, and the predicted time length is 7 days, namely t =7.
The ConvLSTM initial basic parameters are: the filter =16 adopted by the first layer, the convolution kernel size is 3 × 3, and the span strides adopts 1 × 1; the second layer adopts filters =32, the convolution kernel size is 7 × 7, and strides adopts 3 × 3;
and after inputting the given daily oscillation record number, predicting to obtain an oscillation disturbance quantity form chart which is possible to occur on the next day by the trained prediction module.
And 3.2, specially, adding an attention mechanism layer as output after the full connection layer.
For temporal prediction, the input time is in units of weeks, so there is an implied cycle in the corresponding input 3D image time sequence also in "weeks" time span; thus, when training and predicting future week data, the particular days predicted within a week are most relevant to the particular day corresponding to the input information, e.g., monday for monday, sunday for sunday. Thus improving the Convlstm grid structure and adding a softmax layer after the full link layer. So as to realize attention training and prediction under the cycle of the week.
The Softmax function is defined as follows:
Wherein
C is the number of output nodes (i.e. the number of output nodes of the full connection layer), i.e. the number of classes to be classified.
Finally, the invention also provides a system, which comprises
A first module: the node relation matrix is configured to draw an electrical node virtual plane map based on the node relation matrix, and specifically, the node relation matrix is modified circularly and equivalently to enable a three-dimensional power grid network node structure to be equivalent to an equivalent plane power grid network node structure;
a second module: the system comprises a network node structure, a seasonal adaptive CNN network model and a power grid, wherein the network node structure is configured to acquire oscillation source data of an equivalent plane power grid network node structure, preprocess the oscillation source data and output the oscillation source data to train the oscillation source data in the seasonal adaptive CNN network model so as to acquire oscillation source classification data suitable for setting different seasonal periods of a power grid;
a third module: the method is configured to construct a ConvLSTM network-based oscillation source risk analysis space-time distribution prediction model, the prediction model is trained and tested by dividing oscillation source classification data into a training set and a testing set, the optimized prediction model is obtained, and the prediction model can predict and obtain an oscillation disturbance quantity form graph which is possible to occur in the next day after a given daily oscillation record number is input.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.
Claims (8)
1. A regional power grid oscillation source risk prediction and control method based on a convolution long-short term memory neural network is characterized by comprising the following steps:
drawing an electrical node virtual plane map based on the node relation matrix, specifically, equating a three-dimensional power grid network node structure to an equivalent plane power grid network node structure by circularly and equivalently modifying the node relation matrix;
in the case of an equivalent conversion, the conversion is carried out,
determining a plurality of 500 kilovolt electrical nodes to which any one 110 kilovolt electrical node is subordinate to form a matrix T _5_1, wherein the matrix T _5_1 is a 3 × N _500matrix;
traversing a given 110KV electrical node, inquiring non-0 elements of a corresponding row of a node i in an initial node relation matrix, and deleting the row except for a diagonal array;
determining pairs of adjacent 500kV electrical nodes by using a matrix T _5_1, and determining matrix element values corresponding to the associated electrical node nodes by using the following formula until all 110kV electrical nodes are traversed;
wherein the content of the first and second substances,
；
acquiring oscillation source data of an equivalent plane power grid node structure, preprocessing the oscillation source data, outputting the preprocessed oscillation source data to a seasonal adaptive CNN network model, and training the preprocessed oscillation source data to obtain oscillation source classification data suitable for different seasonal periods of a set power grid;
constructing an oscillation source risk analysis space-time distribution prediction model based on a ConvLSTM network, dividing oscillation source classification data into a training set and a testing set to train and test the prediction model to obtain an optimized prediction model, wherein the prediction model can predict an oscillation disturbance quantity form chart which is possible to occur in the second day after inputting a given daily oscillation record number in the current day, and specifically comprises the following steps:
through four parts of ConvLSTM network design, relevant environmental impact factor fusion and prediction result evaluation method, a regional power grid oscillation source space-time distribution prediction model based on the ConvLSTM network is constructed, a 3D matrix with an image matrix L x L is created, and one element is arranged in any t dimension of the 3D geographic matrix with a 3D matrix C structure L x D
,
The numerical value of the node is represented as the number of single-day oscillation amplitude values recorded by the node, t is a time dimension, 1 day is a time scale unit, and the predicted time length is 7 days, namely t =7;
the ConvLSTM initial basic parameters are: the filter =16 adopted by the first layer, the convolution kernel size is 3 × 3, and the span strides adopts 1 × 1; the second layer used filters =32, convolution kernel size 7 × 7, and strides used 3 × 3.
2. The method for predicting and controlling the risk of the oscillation source of the regional power Grid based on the convolutional long and short term memory neural network as claimed in claim 1, wherein the Grid node association admittance matrix of the electric network is a _ Grid node number N:
for arbitrary
And the electric association relation between the node i and the node j is represented, and the row and column ranges of the node association admittance matrix can be configured to set the classified kilovolt nodes.
3. The method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolutional long-short term memory neural network as claimed in claim 2, wherein the row-column range is specifically configured as follows:
all 500kV nodes are arranged in a line from 1-N _500 to 200 kV electrical nodes connected with 220kV and 500kV from N _500+1 to N_220 _500, from N _220_500+1 to N _220 _220are common nodes of 220kV, and from N _220+1 to N are common nodes of 110 kV; the diagonal element defaults to 1; other principles for arbitrary
And when a direct electric line link exists between the node i and the node j, setting the value to be 1, otherwise, setting the value to be 0.
4. The method for predicting and controlling the risk of the regional power grid oscillation source based on the convolution long-short term memory neural network as claimed in claim 2, wherein the CNN network model comprises an oscillation preprocessing layer, the oscillation source data is preprocessed, and image compression is performed after daily disturbance pattern scanning is performed by adopting an activation function, specifically, the image compression is performed
Creating a 2D geographic matrix with an image matrix of M x M, wherein the 2D geographic matrix is in single mapping with a node relation matrix corresponding to a node;
changing the image matrix A _ Grid into A _ oc _ Grid by adopting an improved sigmoid activation function:
wherein, according to the classical theory of forced power oscillation,
wherein A is the maximum amplitude after PRONY decomposition, and
the damping value corresponding to the maximum amplitude after decomposition;
compressing blank invalid areas of the images in the image matrix A _ oc _ Grid, and deleting all blank value rows and columns in the image matrix A _ oc _ Grid to generate A _ oc _ Grid _ dense.
5. The method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolutional long-short term memory neural network as claimed in claim 1, wherein the CNN network model further comprises at least two CONV layers, the input is regional terrain elements, the output is processed regional terrain elements, and the activation function of each layer is an improved relu function:
n is a positive integer and is determined by combining artificial experience and retrograde motion according to different seasons.
6. The method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolutional long-short term memory neural network as claimed in claim 1, wherein an attention mechanism layer is added as an output after a full connection layer in the prediction model, and a Softmax function layer is added after the full connection layer, and the Softmax function layer is defined as follows:
wherein
The number of the output nodes is the number of the full-connection layer output, and the number of the full-connection layer output is the classified category number.
7. The method for predicting and controlling the risk of the oscillation source of the regional power grid based on the convolutional long and short term memory neural network as claimed in claim 1, wherein the input data of the prediction model is a 3D image matrix of L x L, and the corresponding nodes in the 3D matrix and the corresponding nodes in the 2D geographic matrix are in a single mapping.
8. A system, comprising
A first module: the node relation matrix is configured to draw an electrical node virtual plane map based on the node relation matrix, and specifically, the node relation matrix is modified through cyclic equivalence to enable a three-dimensional power grid network node structure to be equivalent to an equivalent plane power grid network node structure;
in the case of an equivalent conversion, the conversion,
determining a plurality of 500 kilovolt electrical nodes to which any one 110 kilovolt electrical node is subordinate to form a matrix T _5_1, wherein the matrix T _5_1 is a 3 × N _500matrix;
traversing a given 110KV electrical node, inquiring a non-0 element in a corresponding column of a node i in an initial node relation matrix, and deleting the column except for a diagonal array;
determining the adjacent 500kV electrical node pairs by using a matrix T _5_1, and determining matrix element values corresponding to the associated electrical node pairs by using the following formula until all 110kV electrical nodes are traversed;
wherein, the first and the second end of the pipe are connected with each other,
；
a second module: the system comprises a network node structure, a seasonal adaptive CNN network model and a power grid, wherein the network node structure is configured to acquire oscillation source data of an equivalent plane power grid network node structure, preprocess the oscillation source data and output the oscillation source data to train the oscillation source data in the seasonal adaptive CNN network model, so that oscillation source classification data suitable for setting different seasonal periods of the power grid are acquired;
a third module: the method is configured to construct a ConvLSTM network-based oscillation source risk analysis space-time distribution prediction model, train and test the prediction model by dividing oscillation source classification data into a training set and a test set to obtain an optimized prediction model, and the prediction model can predict and obtain an oscillation disturbance quantity form chart which is possible to occur in the second day after inputting a given daily oscillation record number in the current day, and specifically comprises the following steps:
through four parts of ConvLSTM network design, relevant environmental impact factor fusion and prediction result evaluation method, a regional power grid oscillation source space-time distribution prediction model based on the ConvLSTM network is constructed, a 3D matrix with an image matrix L x L is created, and one element is arranged in any t dimension of the 3D geographic matrix with a 3D matrix C structure L x D
,
The numerical value of the node is represented by the number of single-day oscillation amplitude values recorded by the node, t is a time dimension, a time scale is 1 day, and the predicted time length is 7 days, namely t =7;
the ConvLSTM initial basic parameters are: the filter filters =16 adopted by the first layer, the convolution kernel size is 3 × 3, and the span strides adopts 1 × 1; the second layer used filters =32, convolution kernel size 7 × 7, and strides used 3 × 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211117058.5A CN115187154B (en) | 2022-09-14 | 2022-09-14 | Neural network-based regional power grid oscillation source risk prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211117058.5A CN115187154B (en) | 2022-09-14 | 2022-09-14 | Neural network-based regional power grid oscillation source risk prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115187154A CN115187154A (en) | 2022-10-14 |
| CN115187154B true CN115187154B (en) | 2022-12-16 |
Family
ID=83524833
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211117058.5A Active CN115187154B (en) | 2022-09-14 | 2022-09-14 | Neural network-based regional power grid oscillation source risk prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115187154B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115577226B (en) * | 2022-12-08 | 2023-03-03 | 南方电网数字电网研究院有限公司 | Forced oscillation source positioning method, device, equipment and storage medium |
| CN115983507B (en) * | 2023-03-20 | 2023-06-16 | 华中科技大学 | Method and system for predicting broadband oscillation risk of section of power grid of transmitting end source |
| CN117110787B (en) * | 2023-08-29 | 2024-02-09 | 燕山大学 | Subsynchronous oscillation source positioning method of quaternary feature set convolutional neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112200038A (en) * | 2020-09-29 | 2021-01-08 | 国网四川省电力公司经济技术研究院 | CNN-based rapid identification method for oscillation type of power system |
| US11164068B1 (en) * | 2020-11-13 | 2021-11-02 | International Business Machines Corporation | Feature recognition with oscillating neural network |
| CN114818483A (en) * | 2022-04-14 | 2022-07-29 | 东南大学溧阳研究院 | Electromechanical disturbance positioning and propagation prediction method based on graph neural network |
| CN115022193A (en) * | 2022-05-23 | 2022-09-06 | 电子科技大学 | Local area network flow prediction method based on improved ConvLSTM deep learning model |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112183368B (en) * | 2020-09-29 | 2024-02-02 | 国网四川省电力公司经济技术研究院 | LSTM-based rapid identification method for low-frequency oscillation modal characteristics of power system |
| US20220115871A1 (en) * | 2020-10-08 | 2022-04-14 | Zhe Yu | Power System Low-Frequency Oscillation Mechanism Identification with CNN and Transfer Learning |
| US11176442B1 (en) * | 2021-02-11 | 2021-11-16 | North China Electric Power University | Fast power system disturbance identification using enhanced LSTM network with renewable energy integration |
-
2022
- 2022-09-14 CN CN202211117058.5A patent/CN115187154B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112200038A (en) * | 2020-09-29 | 2021-01-08 | 国网四川省电力公司经济技术研究院 | CNN-based rapid identification method for oscillation type of power system |
| US11164068B1 (en) * | 2020-11-13 | 2021-11-02 | International Business Machines Corporation | Feature recognition with oscillating neural network |
| CN114818483A (en) * | 2022-04-14 | 2022-07-29 | 东南大学溧阳研究院 | Electromechanical disturbance positioning and propagation prediction method based on graph neural network |
| CN115022193A (en) * | 2022-05-23 | 2022-09-06 | 电子科技大学 | Local area network flow prediction method based on improved ConvLSTM deep learning model |
Non-Patent Citations (1)
| Title |
|---|
| Estimation of dominant power oscillation modes based on ConvLSTM approach using synchrophasor data and cross-validation technique;Fanta Senesoulin1e等;《Sustainable Energy, Grids and Networks》;20220910;正文第1-12页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115187154A (en) | 2022-10-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115187154B (en) | Neural network-based regional power grid oscillation source risk prediction method and system | |
| US20220373984A1 (en) | Hybrid photovoltaic power prediction method and system based on multi-source data fusion | |
| CN109657839B (en) | Wind power prediction method based on deep convolutional neural network | |
| Li et al. | Predicting short-term electricity demand by combining the advantages of arma and xgboost in fog computing environment | |
| CN109190672A (en) | Operation of Electric Systems operating condition unsupervised clustering and device | |
| CN114240000A (en) | Air quality prediction method based on space-time graph convolution network | |
| Zhang et al. | Optimal graph structure based short-term solar PV power forecasting method considering surrounding spatio-temporal correlations | |
| CN107193060A (en) | A kind of multipath Typhoon Storm Surge Over method for quick predicting and system | |
| Wang et al. | Dynamic directed graph convolution network based ultra‐short‐term forecasting method of distributed photovoltaic power to enhance the resilience and flexibility of distribution network | |
| Tian et al. | Wind power forecasting by the BP neural network with the support of machine learning | |
| CN113344283B (en) | Energy internet new energy consumption capability assessment method based on edge intelligence | |
| Chen et al. | Graph neural network-based wind farm cluster speed prediction | |
| Yin et al. | A survey of deep learning and its application in distribution network | |
| Zhang et al. | Joint forecasting of regional wind and solar power based on attention neural network | |
| CN117114081A (en) | Distributed photovoltaic power prediction method and device based on transfer learning | |
| Deng et al. | Hybrid short-term load forecasting using the Hadoop MapReduce framework | |
| Zhang et al. | Graph Neural Network Based Short-Term Solar Irradiance Forcasting Model Considering Surrounding Meteorological Factors | |
| CN115392525A (en) | Distributed wind power prediction method, device and equipment | |
| Jiang et al. | Transmission line modeling based on 3D laser scanning point cloud | |
| Ge et al. | Virtual Collection for Distributed Photovoltaic Data: Challenges, Methodologies, and Applications. Energies 2022, 15, 8783 | |
| CN113408191A (en) | PM2.5 prediction method based on graph self-supervision learning and storage medium | |
| Liu et al. | SS-IPLE: Semantic segmentation of electric power corridor scene and individual power lines extraction from UAV-based Lidar point cloud | |
| Nguyen Thanh et al. | Insulator leakage current prediction using hybrid of particle swarm optimization and gene algorithm-based neural network and surface spark discharge data | |
| Shen et al. | Application of digital twin technology in auxiliary decision-making system for grid-connected dispatching of new energy | |
| Wang et al. | Analysis of the spatio-temporal characteristics of intercity travel based on SVD and complex network: take Bohai Rim City Group as an example |
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 |