CN115275990A - Evaluation method and system for broadband oscillation risk of regional power grid - Google Patents
Evaluation method and system for broadband oscillation risk of regional power grid Download PDFInfo
- Publication number
- CN115275990A CN115275990A CN202210903376.8A CN202210903376A CN115275990A CN 115275990 A CN115275990 A CN 115275990A CN 202210903376 A CN202210903376 A CN 202210903376A CN 115275990 A CN115275990 A CN 115275990A
- Authority
- CN
- China
- Prior art keywords
- power grid
- data
- neural network
- regional power
- deep
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000010355 oscillation Effects 0.000 title claims abstract description 145
- 238000011156 evaluation Methods 0.000 title claims description 11
- 238000000034 method Methods 0.000 claims abstract description 49
- 230000002159 abnormal effect Effects 0.000 claims abstract description 41
- 230000006870 function Effects 0.000 claims description 92
- 238000013527 convolutional neural network Methods 0.000 claims description 48
- 238000013528 artificial neural network Methods 0.000 claims description 38
- 238000012549 training Methods 0.000 claims description 36
- 230000004913 activation Effects 0.000 claims description 34
- 239000011159 matrix material Substances 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 20
- 238000003860 storage Methods 0.000 claims description 16
- 238000005259 measurement Methods 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 13
- 230000003044 adaptive effect Effects 0.000 claims description 12
- 238000007781 pre-processing Methods 0.000 claims description 11
- 238000004088 simulation Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 230000002452 interceptive effect Effects 0.000 claims 2
- 230000007246 mechanism Effects 0.000 abstract description 11
- 238000011176 pooling Methods 0.000 description 22
- 238000010586 diagram Methods 0.000 description 14
- 238000012545 processing Methods 0.000 description 13
- 238000004458 analytical method Methods 0.000 description 8
- 230000003993 interaction Effects 0.000 description 8
- 210000002569 neuron Anatomy 0.000 description 8
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000012502 risk assessment Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000000295 complement 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
- 238000013016 damping Methods 0.000 description 1
- 230000008846 dynamic interplay Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- 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
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/002—Flicker reduction, e.g. compensation of flicker introduced by non-linear load
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/10—Power transmission or distribution systems management focussing at grid-level, e.g. load flow analysis, node profile computation, meshed network optimisation, active network management or spinning reserve management
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- Power Engineering (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Entrepreneurship & Innovation (AREA)
- Primary Health Care (AREA)
- General Health & Medical Sciences (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Nonlinear Science (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses an assessment method and system for broadband oscillation risk of a regional power grid, and belongs to the technical field of power system automation. The method comprises the following steps: acquiring historical data of abnormal fluctuation of a regional power grid, and acquiring a deep convolution model based on the historical data; acquiring abnormal fluctuation data of a target regional power grid, inputting the abnormal fluctuation data serving as risk prediction data into the deep convolution model, and identifying a broadband oscillation mode of the regional power grid based on the deep convolution model; collecting the broadband oscillation mode data of the regional power grid identified by the deep convolution model to obtain a broadband oscillation set, and determining the broadband oscillation risk of the target regional power grid according to the broadband oscillation set. The broadband oscillation identified by the method is high in precision, and uncertainty caused by a plurality of influence factors based on a mechanism model is avoided.
Description
Technical Field
The present invention relates to the field of power system automation technology, and more particularly, to a method and a system for evaluating a risk of broadband oscillation of a regional power grid.
Background
The traditional power system oscillation mainly comprises low-frequency oscillation and SSR/SSO, and the modern double-high power system causes novel oscillation with the characteristics of broadband time-varying characteristic, strong nonlinearity and the like due to interaction between power electronic equipment and various elements of a power grid. Unlike the traditional power system oscillation mechanism, the novel oscillation is mainly electromagnetic oscillation caused by power electronic control, and the frequency range relates to 10 < -1 > to 103Hz, so the novel oscillation is called 'broadband oscillation'. Traditionally, traditional oscillation and novel broadband oscillation are subdivided into low-frequency oscillation (0.1-2.5 Hz), subsynchronous oscillation (several Hz-2 times power frequency) and medium-high frequency oscillation (hundreds Hz to thousands Hz) according to the different frequency intervals. The traditional low-frequency oscillation is dominated by the swing characteristic among the rotors of the generator in a disturbance system, and the forming mechanisms comprise a negative damping mechanism, a forced oscillation mechanism, a chaos mechanism, a parameter resonance mechanism and the like. In addition, the oscillation caused by the renewable energy source unit or the current transformer phase-locked loop due to improper parameters of the electromechanical scale controller can also fall into the frequency range of low-frequency oscillation, and is called electromechanical-like low-frequency oscillation. According to the definition of IEEE, the conventional subsynchronous oscillation is generally divided into two categories, namely SSR generated by coupling of a steam turbine generator unit and a series compensation system and SSO generated by interaction of the steam turbine generator unit and a fast controller. The novel subsynchronous oscillation caused by the converter type power grid connection of wind power and the like is essentially different from the traditional subsynchronous oscillation, and mainly relates to the dynamic interaction of the converter control of a wind turbine generator and a series compensation system or a weak alternating current power grid in the mechanism, which is generally called subsynchronous control interaction (SSCI). In a new energy power system, when subsynchronous power oscillation occurs, voltage and current harmonics often include a supersynchronous component complementary to the subsynchronous component in addition to the subsynchronous component. Because various generator sets, series compensation, rapid controllers and the like exist simultaneously in an actual power grid, the various oscillation forms usually coexist, and are mutually coupled and influenced, so that complex and various subsynchronous/supersynchronous oscillation phenomena are formed.
In the past, classical deep convolutional neural network models include a LeNet model, an AlexNet model, a VGG-Net model, a GoogLeNet model and a ResNet model. With the gradual deepening of the application field, the deep convolutional neural network has been developed greatly, and the related applications of image processing and target identification are comprehensively promoted by the rapid development of computer hardware and software, the establishment of large-scale data sets and the breakthrough of related technical fields. The convolution neural network is used as a comprehensive discipline which integrates computer science, statistics, cranial neurology and sociology, and promotes computers to have intelligence as common as human beings. In many dangerous works and enterprise production, the convolutional neural network can replace workers to realize functions of identification, judgment, analysis and the like. At present, artificial intelligence is widely popularized, and recognition accuracy of image problems is greatly improved by means of machine vision, deep learning and other advanced technologies.
In a novel electric power system, a large amount of new energy systems are gradually connected into a regional power grid, namely, the risk degree of broadband oscillation is increased day by day, the power processing of the new energy systems is greatly influenced by weather factors, and in addition, the starting and stopping of a traditional unit and the gradual increase of overhauling flexibility are increased, the risk of broadband large-area oscillation of the power grid is gradually increased, and the seasonal broadband oscillation risk assessment of the regional power grid gradually becomes one of core problems concerned by the operation of the electric power system. Seasonal broadband oscillation risk assessment and prejudgment are conducted on the regional power grid under the characteristics of the novel power system, so that power system operators can assess the risk degree of broadband oscillation occurring on the power grid in a certain future interval in advance, optimization processing is conducted on the starting mode and the maintenance mode of each sub-region of the regional power grid by combining assessment results, and the probability of large-area broadband oscillation is reduced.
Since the systematic generation of broadband oscillation and the propagation mechanism in the power grid are not clear, many researchers have conducted relevant research from the viewpoint of mechanical analysis, but the mechanism analysis is still in a highly developed and discussed stage due to the large-area rapid equipment access system with broadband oscillation generation source. Particularly, in the aspect of overall risk assessment of a regional power grid, the risk assessment method is limited by the existing simulation means, and in the aspect of power grid operation control, rapid risk assessment is realized through the simulation means at the early stage, and a certain time is still needed. On the other hand, with the development of deep learning computer technology and the large-area popularization of power grid sensor-PMU configuration, power grid oscillation analysis based on data driving gradually becomes one of solutions for power grid oscillation evaluation. Firstly, for any given regional power grid, due to the fact that certain priori knowledge is provided, such as a power grid topological structure, a new energy system access electrical node, local high-precision PMU measurement, starting and stopping of a traditional thermal power generating unit and other information, a certain priori data accumulation is provided for a power grid oscillation characteristic rule of a basic topological structure, secondly, due to the rapid development of a convolutional neural network in the technical field of deep learning, a data structure and a calculation framework structure of the convolutional neural network on a driving problem are rapidly developed in recent years, and the convolutional neural network becomes possible in processing broadband oscillation panoramic analysis of a relatively fixed topological power grid.
Disclosure of Invention
In order to solve the above problem, the present invention provides an evaluation method for broadband oscillation risk of a regional power grid, including:
acquiring historical data of abnormal fluctuation of a regional power grid, and acquiring a deep convolution model based on the historical data;
acquiring abnormal fluctuation data of a target regional power grid, inputting the abnormal fluctuation data serving as risk prediction data into the deep convolution model, and identifying a broadband oscillation mode of the regional power grid based on the deep convolution model;
collecting the broadband oscillation mode data of the regional power grid identified by the deep convolution model to obtain a broadband oscillation set, and determining the broadband oscillation risk of the target regional power grid according to the broadband oscillation set.
Optionally, the acquiring historical data of the abnormal fluctuation of the regional power grid, and acquiring a deep convolution model based on the historical data includes:
acquiring historical data of abnormal fluctuation of a regional power grid, preprocessing the historical data, acquiring multi-frequency amplitude data of two side points of a PMU (phasor measurement Unit) in the same time scene, and converting the multi-frequency amplitude data into a multi-channel color map;
and taking the multi-channel color image as input data of a pre-built deep convolution neural network, training the deep convolution neural network, and obtaining a deep convolution model.
Optionally, the preprocessing is performed on the historical data, and the multi-frequency amplitude data of the PMU two side points in the same time scene are obtained, including: determining abnormal fluctuation of a power grid as large fluctuation caused by non-power grid line short circuit tripping faults or unit tripping faults according to historical data, establishing a tensor for an oscillation scene at a single moment in the historical data, establishing a multi-dimensional matrix for a given power grid area, and acquiring multi-frequency amplitude data of PMU (phasor measurement unit) points at two sides in the same moment scene according to the tensor and the multi-dimensional matrix.
Optionally, the pre-established deep convolutional neural network includes an activation function;
the activation function, as follows:
wherein g (x) is an activation function, e is a constant, the function value of g (x) is 0 when x <0, and the gradient is 0, and a Tanh function is output when x > 0.
Optionally, the pre-constructed convolutional neural network further includes: a neural network architecture;
the convolution kernel of the convolutional layer of the neural network architecture is as follows:
wherein Y isnFor the nth output tensor, xiIn order to input the tensor,
for the nth convolution kernel on the ith channel, bnIs the offset, and M is a constant of the total number of channels.
Optionally, training the deep convolutional neural network includes: adjusting dynamic hyper-parameters of the deep convolutional neural network;
the dynamic hyper-parameters comprise: epoch and batch parameters, loss function, and learning rate.
Optionally, when the deep convolutional neural network is trained, the optimal solution is searched in a multi-process multi-starting point parallel search interaction manner.
Optionally, a full electromagnetic simulation network is used to collect the broadband oscillation mode data.
Optionally, when the deep convolution model identifies the broadband oscillation mode of the regional power grid, if the identification fails or the identification degree does not meet the requirement, the oscillation source is locked, the broadband oscillation mode is determined according to the oscillation source, and the determined broadband oscillation mode is loaded into the stack of the deep convolution model as a new mode.
The invention also provides an evaluation system for the broadband oscillation risk of the regional power grid, which comprises the following steps:
the training unit is used for acquiring historical data of abnormal fluctuation of a regional power grid and acquiring a deep convolution model based on the historical data;
the output unit is used for acquiring abnormal fluctuation data of a target area power grid, inputting the abnormal fluctuation data serving as risk prediction data into the deep convolution model, and identifying a broadband oscillation mode of the area power grid based on the deep convolution model;
and the evaluation unit is used for collecting the broadband oscillation mode data of the regional power grid identified by the deep convolution model, acquiring a broadband oscillation set and determining the broadband oscillation risk of the target regional power grid according to the broadband oscillation set.
Optionally, acquiring historical data of abnormal fluctuation of the regional power grid, and acquiring a deep convolution model based on the historical data includes:
acquiring historical data of abnormal fluctuation of a regional power grid, preprocessing the historical data, acquiring multi-frequency amplitude data of two side points of a PMU (phasor measurement Unit) in the same time scene, and converting the multi-frequency amplitude data into a multi-channel color map;
and taking the multi-channel color image as input data of a pre-built deep convolution neural network, training the deep convolution neural network, and obtaining a deep convolution model.
Optionally, the training unit preprocesses the historical data to obtain PMU two-side-point multi-frequency amplitude data in the same time scene, including: determining abnormal fluctuation of a power grid as large fluctuation caused by non-power grid line short circuit tripping faults or unit tripping faults according to historical data, establishing a tensor for an oscillation scene at a single moment in the historical data, establishing a multi-dimensional matrix for a given power grid area, and acquiring multi-frequency amplitude data of PMU (phasor measurement unit) points at two sides in the same moment scene according to the tensor and the multi-dimensional matrix.
Optionally, the deep convolutional neural network pre-established in the training unit includes an activation function;
the activation function, as follows:
wherein g (x) is an activation function, e is a constant, the function value of g (x) is 0 when x <0, and the gradient is 0, and a Tanh function is output when x > 0.
Optionally, the convolutional neural network pre-built in the training unit includes: a neural network architecture and activation function;
the convolution kernel of the convolutional layer of the neural network architecture is as follows:
wherein Y isnFor the nth output tensor, xiIn order to input the tensor,
for the nth convolution kernel on the ith channel, bnIs offset, M is a constant of the total number of channels;
the activation function, as follows:
wherein g (x) is an activation function, e is a constant, the function value of g (x) is 0 when x <0, and the gradient is 0, and a Tanh function is output when x > 0.
Optionally, the training unit trains the deep convolutional neural network, including: adjusting dynamic hyper-parameters of the deep convolutional neural network;
the dynamic hyper-parameters comprise: epoch and batch parameters, loss functions, and learning rates.
Optionally, when the training unit performs training on the deep convolutional neural network, the optimal solution is searched in a multi-process multi-starting-point parallel search interaction manner, where the method includes:
searching an optimal solution by using N +1 CPUs, controlling each CPU to start iterative computation from an initial solution with the N being greater than 10, obtaining a computation result, and sending an adaptive value of the computation result to a unified server to be ordered according to fitness;
after the sorting is finished, selecting an operator and sorting according to the adaptive value by adopting a selection algorithm, and if the calculation result of the (N + 1) th CPU is not superior to the calculation results of the N CPUs, randomly allocating the most worried solution of the iterative calculation to the N CPUs;
and controlling the N +1 CPUs to continuously execute iterative computation, and stopping the iterative computation when the loss value is lower than the threshold value.
Optionally, the output unit collects the broadband oscillation mode data by using a full electromagnetic simulation network.
Optionally, when the output unit identifies the broadband oscillation mode of the regional power grid through the deep convolution model, if the identification fails or the identification degree does not meet the requirement, the oscillation source is locked, the broadband oscillation mode is determined according to the oscillation source, and the determined broadband oscillation mode is loaded into the stack of the deep convolution model as a new mode.
In yet another aspect, the present invention further provides a computing device comprising: one or more processors;
a processor for executing one or more programs;
when the one or more programs are executed by the one or more processors, the method for evaluating the risk of broadband oscillation of the regional power grid is realized.
In still another aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed, implements a method for assessing risk of broadband oscillation of a regional power grid as described above.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides an assessment method and system for regional power grid broadband oscillation risk.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of a method Dropout of the present invention;
FIG. 3 is a graph of activation function output for the method of the present invention;
FIG. 4 is a diagram of a deep convolutional neural network of the present invention;
fig. 5 is a block diagram of the system of the present invention.
Detailed Description
The exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, however, the present invention may be embodied in many different forms and is not limited to the embodiments described herein, which are provided for complete and complete disclosure of the present invention and to fully convey the scope of the present invention to those skilled in the art. The terms used in the exemplary embodiments shown in the drawings are not intended to limit the present invention. In the drawings, the same unit/element is denoted by the same reference numeral.
Unless otherwise defined, terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Further, it will be understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense.
Example 1:
the invention provides an evaluation method for broadband oscillation risk of a regional power grid, as shown in fig. 1, comprising the following steps:
s1, acquiring historical data of abnormal fluctuation of a regional power grid, and acquiring a deep convolution model based on the historical data;
s2, acquiring abnormal fluctuation data of a target area power grid, inputting the abnormal fluctuation data serving as risk prediction data into the deep convolution model, and identifying a broadband oscillation mode of the area power grid based on the deep convolution model;
and S3, collecting the broadband oscillation mode data of the regional power grid identified by the deep convolution model, acquiring a broadband oscillation set, and determining the broadband oscillation risk of the target regional power grid according to the broadband oscillation set.
The method comprises the following steps of collecting historical data of abnormal fluctuation of a regional power grid, and acquiring a deep convolution model based on the historical data, wherein the method comprises the following steps:
acquiring historical data of abnormal fluctuation of a regional power grid, preprocessing the historical data, acquiring multi-frequency amplitude data of two side points of a PMU (phasor measurement Unit) in the same time scene, and converting the multi-frequency amplitude data into a multi-channel color map;
and taking the multi-channel color image as input data of a pre-built deep convolution neural network, training the deep convolution neural network, and obtaining a deep convolution model.
Preprocessing the historical data to obtain multi-frequency amplitude data of PMU points at two sides in the same time scene, wherein the preprocessing comprises the following steps: determining abnormal fluctuation of a power grid as large fluctuation caused by non-power grid line short circuit tripping faults or unit tripping faults according to historical data, establishing a tensor for an oscillation scene at a single moment in the historical data, establishing a multi-dimensional matrix for a given power grid area, and acquiring multi-frequency amplitude data of PMU (phasor measurement unit) points at two sides in the same moment scene according to the tensor and the multi-dimensional matrix.
The pre-built deep convolutional neural network comprises an activation function;
the activation function, as follows:
wherein g (x) is an activation function, e is a constant, the function value of g (x) is 0 when x <0, and the gradient is 0, and a Tanh function is output when x > 0.
Wherein, the deep convolution neural network of buildding in advance includes: a neural network architecture;
the convolution kernel of the convolutional layer of the neural network architecture is as follows:
wherein Y isnFor the nth output tensor, xiIn order to input the tensor,
for the nth convolution kernel on the ith channel, bnIs the offset, and M is a constant of the total number of channels.
Wherein training the deep convolutional neural network comprises: adjusting dynamic hyper-parameters of the deep convolutional neural network;
the dynamic hyper-parameters comprise: epoch and batch parameters, loss functions, and learning rates.
When the deep convolutional neural network is trained, searching for an optimal solution in a multi-process multi-starting point parallel search interaction mode comprises the following steps:
searching an optimal solution by using N +1 CPUs, controlling each CPU to start iterative computation from an initial solution with the N being greater than 10, obtaining a computation result, and sending an adaptive value of the computation result to a unified server to be ordered according to fitness;
after the sorting is finished, operators are selected and sorted according to the adaptive values by adopting a selection algorithm, and if the calculation result of the (N + 1) th CPU is not superior to the calculation results of the N CPUs, the most worrisome solution of the iterative calculation is randomly distributed to the N CPUs;
and controlling the N +1 CPUs to continuously execute iterative computation, and stopping the iterative computation when the loss value is lower than the threshold value.
And the full electromagnetic simulation network collects the broadband oscillation mode data.
When the deep convolution model identifies the broadband oscillation mode of the regional power grid, if the identification fails or the identification degree does not meet the requirement, the oscillation source is locked, the broadband oscillation mode is determined according to the oscillation source, and the determined broadband oscillation mode is used as a new mode to be installed into the stack mode of the deep convolution model.
In step S1, preprocessing the historical data to obtain multi-frequency amplitude data of PMU two side points in the same time scene includes: determining abnormal fluctuation of a power grid as large fluctuation caused by non-power grid line short circuit tripping faults or unit tripping faults according to historical data, establishing a tensor for an oscillation scene at a single moment in the historical data, establishing a multi-dimensional matrix for a given power grid area, and acquiring multi-frequency amplitude data of PMU (phasor measurement unit) points at two sides in the same moment scene according to the tensor and the multi-dimensional matrix.
The method comprises the following specific steps:
firstly, based on a D5000 system, preliminarily analyzing various collected historical data such as abnormal fluctuation of a massive regional power grid and the like by manpower to know the data to a certain extent, eliminating large fluctuation caused by short circuit trip fault of a power grid line and unit trip fault, then establishing a tensor (tensor) by the data in an oscillation scene at a single moment, and setting a given regional power grid as the given regional power grid
A matrix of dimensions, (N × 1 × 2.. M), wherein:
(1) L and N are coordinates of the electrical nodes which are obtained after the grid is regarded as two-dimensional gridding;
(2)
m given interest for a given gridmaxAnd +1 frequency oscillation frequency band.
(for example, in a given regional power grid, only 0.1Hz-100Hz wide frequency oscillations are observed at a node, and M is 1 Hz-2 Hz1The oscillation conditions from the ith Hz to the (i + 1) th Hz are stored in Mi)
For each frequency band, the frequency bands may be divided according to the actual conditions based on the actual conditions of a given grid.
(3) Meanwhile, the disturbance amplitude characteristic coefficient of the regional power grid is as follows: at a given time, carrying out PRONY analysis on the disturbance of a given electrical geographic coordinate to obtain the frequency spectrum analysis of the disturbance waveform record at the given time, and putting the frequency into a given dimensional coordinate of a given geographic electrical node in the tensor according to the obtained evaluation analysis. In general, the amplitude can be expressed as: a = (A1, A2,. An), n is the maximum number of frequency bands available by Prony analysis.
In step 1, the construction of the deep convolutional neural network comprises the construction of a protective neural network architecture and an activation function, and the core of the convolutional neural network is the construction of a convolutional layer and a pooling layer.
Convolutional layers, as follows:
convolution kernel background, as follows:
the convolution layer is the most key core part of the convolutional neural network, the implementation process is that the convolution kernel with a certain size is used for weighting and summing an input signal matrix by an area with the same size, the process is equivalent to the process of multiplying, accumulating and summing all neuron data and connected weights in the artificial neural network, and the implementation function is to extract the characteristics of a received signal matrix through convolution calculation and map the characteristics to the next layer of output characteristics. The convolution operation is a linear operation method with translation invariance, and directly shows the properties of signals.
The mathematical expression for the convolution operation is given below:
assuming that there are two consecutive integrable functions f (t) and g (t), if the following relationship exists:
it is called the convolution of a continuous function. Convolution is the result of multiplying the function f (t) and the function g (t) in a certain number domain range and then summing, and is specifically explained as that one function is overturned, is continuously moved (sliding window scanning) in a specified numerical range, is multiplied by the other function, and then is accumulated and summed to obtain a result;
the convolution operation discrete expression is as follows:
the matrix pattern of the above formula is mathematically described as:
s(n)＝(F*G)(n)
the two-dimensional mathematical expression for the convolution operation is as follows:
the design of the convolution kernel is as follows:
the convolution layer can output different output tensors by performing operations on different convolution kernels and input tensors by forward propagation, and the convolution calculation is as follows.
Wherein, YnFor the nth output tensor, xiIn order to input the tensor,
for the nth convolution kernel on the ith channel, bnIs offset, M is a constant of the total number of channels;
pooling layer, as follows:
pooling is an operation to reduce the space in the height and length directions, and pooling layers generally include an average pooling layer, which is to calculate the average value of a target area, and a maximum pooling layer, which is to take the maximum value out of the target area, and the present invention employs the maximum pooling layer, which has the following characteristics: the method has the advantages that parameters to be learned do not exist, the number of input and output data channels subjected to pooling operation does not change, and robustness is provided for tiny position changes.
Full link layer (with Dropout), as follows:
each node in the fully connected layer is connected to all nodes in the previous layer for mapping the extracted feature representation to the label space of the sample. Because of the fully connected nature of its nodes, the parameters of the fully connected layer are also typically the most.
The convolutional neural network usually adds one to two full-connection layers after the convolutional layer and the pooling layer, the connection mode of the full-connection layer and the connection mode of the layer above the full-connection layer are similar to that of the artificial neural network and are all completely connected, which also illustrates the necessity of the pooling layer, the full-connection layer is used for integrating the local information which is learned by the convolutional layer and the pooling layer and can distinguish the local information, and mapping the information to a sample mark space, and the CNN reduces the occurrence of the over-fitting phenomenon in the neural network by introducing a "Dropout" technology, as shown in fig. 2, a Dropout processing schematic diagram is shown, that is, neurons are hidden with random probability in the neural network training, and when the neural network is trained and updated again, the neuron nodes are discarded.
The activation function, as follows:
because the convolution operation is linear and invariant to data processing, and the trained linear model is not satisfied with classifying real-world complex data, if only linear relation is calculated, no matter how deep the number of layers of the neural network is, the output of the neural network is obtained by linear operation of input data, and the neural network does not have strong classifying capability. The activation function is thus added so that the network model has a capability of approaching non-linear processing. The activation function is also an extremely important component in the network model, and the nonlinear characteristic is added to enable the model to have stronger expression capability on data.
The initial design concept of the activation function refers to a biological neuron model, the biological neuron sets a decision threshold, when the value of an input signal is larger than the threshold, the neuron is started, otherwise, the neuron is not started, the working mechanism of the activation function is consistent with the process, and the Sigmoid, tanh and ReLU functions are very common activation functions.
(a) Sigmoid function, which is once the most commonly used activation function in neural networks, its mathematical expression is as follows:
as shown in the first graph of fig. 3, the output value of the function obtained in the region of 0 to 1 has the characteristics of continuous, uninterrupted, monotonic increase and nonlinearity, and is a good choice for judging and classifying the neural network data, but it has a certain disadvantage, and when x tends to be at two ends of the value range of the independent variable, the variation range of the output function value at the two ends of the independent variable is very small, and the problem of disappearance of the gradient is easy to occur. Meanwhile, the average value of the function output is not 0, which may cause the oscillation convergence rate of the target function to be reduced.
(b) The Tanh function is modified in the Sigmoid function so that its shape is consistent with the Sigmoid function, and the mathematical expression is as follows:
as shown in the second graph of fig. 3, it can be obtained that the output value of the function is in the range of-1 to 1, and the change makes the average value of the output of the function be 0, so that the convergence rate of the target function can be improved. But the problem of gradient disappearance is not well solved.
(c) 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:
g(x)＝max(0,x)
as shown in the third graph of fig. 3, it can be obtained that the output value of the function is in the region of 0 to infinity, which is essentially a comparative piecewise function, with the function value being 0 and the gradient being 0 when x <0, and the function output value being x and the gradient being 1 when x > 0. The ReLU function perfectly circumvents the gradient vanishing phenomenon. However, when the input value of the neuron is less than 0, the corresponding weight will not be updated again, and unfortunately, the output of the ReLU function is not centered on 0 — the distribution of the input data is changed, and the distribution of the next layer of input data is different from the distribution of the previous layer of input data, which greatly reduces the training speed of the model because different input distributions need to be continuously adapted, thereby slowing down the training speed.
The activation function is as follows:
when x is less than 0, the function value is 0 and the gradient is 0, so that the gradient disappearance phenomenon is avoided; when x is larger than 0, the Tanh function is output, so that the output value of the function is in the range of 0 to 1, and the convergence rate is improved.
Output layer, as follows:
the purpose of the output layer is to output the results of the classification and prediction data. Meanwhile, the parameters of each layer are updated by continuously iterating and feeding back the errors, and the loss function is calculated to construct a model with better performance.
The convolutional neural network adopted by the invention is composed of 4 convolutional layers, 4 pooling layers and 2 full-connection layers, as shown in fig. 4, wherein one convolutional layer and one pooling layer are combined into one layer in the neural network, 8 convolutional layers with the size of 5 × 5 in the first layer carry out convolution operation on the characteristic matrix vector of an input signal image, then the result of the convolution operation is delivered to the pooling layer of the same layer for sampling operation, the pooling kernel size of the pooling layer of the first layer is 2 × 2, and the moving step length is 2 × 2. The convolution layer in the second layer uses 16 convolution cores with the size of 5 multiplied by 5 to carry out convolution operation on the mapping characteristic diagram input by the first layer, and then the result of the convolution operation is handed to the pooling layer of the same layer for sampling operation. The third layer and the fourth layer have the same structure scale, the convolution operation is carried out on the mapping characteristic diagram of the input of the previous layer by using 16 convolution cores with the size of 3 multiplied by 3, and the sizes of the pooling layers are the same as those of the first two layers. The fifth and sixth layers are all fully connected layers, using Relu activation function, where the step size of movement for all convolutional layers is 1 × 1, and the pooling layer uses maximum pooling.
The convolutional neural network firstly preprocesses input image data, transmits a data matrix to a convolutional layer for convolution operation, and achieves different feature extraction effects by setting the size of a convolutional kernel. And then, the extracted feature matrix is used as processing data of a pooling layer, and the pooling layer reduces the size of the data through a sampling strategy to inhibit the occurrence of an overfitting phenomenon. And the full connection layer receives the processed characteristic vectors, performs expansion statistics, and outputs a result after the judgment of an activation function.
Output layer, as follows:
the output layer adopts the currently popular softmax output layer.
In step 1, training the deep convolutional neural network includes: adjusting dynamic hyper-parameters of the deep convolutional neural network;
the dynamic hyper-parameters comprise: epoch and batch parameters, loss function, and learning rate.
The array is a 1-dimensional array of 1 x 4;
the fitness function is: the final loss value after each network training is completed;
when the deep convolutional neural network is trained, the optimal solution is searched in a multi-process multi-starting point parallel search interaction mode.
And setting the CPUs of the computers as N +1 (N is more than 10), wherein each CPU is responsible for starting calculation from an initial state, performing M iterations each time, sending the adaptive value to the unified server for comparison and selection after the iteration is completed, and sorting according to the adaptive degree.
And (4) selecting an operator, and adopting a selection algorithm to judge whether the percentage (the percentage is less than 100, and the percentage is manually set and adjusted according to experience) before ranking is superior to the N CPUs in fitness or not. If not, randomly distributing the most solutions of the iteration to percentage computers in N CPUs; the N +1 computers continue to perform computing operations under independent conditions.
And repeating the iteration, wherein the terminal condition is that the loss is lower than the threshold value.
And collecting the broadband oscillation mode data by adopting a full electromagnetic simulation network.
When the deep convolution model identifies the broadband oscillation mode of the regional power grid, if the identification fails or the identification degree does not meet the requirement, the oscillation source is locked, the broadband oscillation mode is determined according to the oscillation source, and the determined broadband oscillation mode is used as a new mode to be installed into a stack mode of the deep convolution model.
The method comprises the following specific steps:
firstly, setting: the image deviation threshold value calculation method comprises the following steps: and calculating the similarity of the plane geometric figure by adopting the improved three-dimensional Euclidean distance. Namely:
d＝a*sprt[(x1-x2)^2+(y1-y2)^2+(z1-z2)^2]
a determination principle: and the distance between the position where the maximum amplitude is located and the position where the maximum amplitude is located is judged, and normalization is carried out. One of the thresholds is manually adjusted.
Extracting an actual oscillation mode; injecting the convolutional neural network pattern recognizer formed in the third step;
judging whether the threshold value is exceeded or not, if not, ignoring, and still using the mode identifier;
if so, the pattern is appended to the new pattern type.
Example 2:
the present invention further provides an evaluation system 200 for evaluating the risk of broadband oscillation of a regional power grid, as shown in fig. 5, including:
the training unit 201 is used for acquiring historical data of regional power grid abnormal fluctuation and acquiring a deep convolution model based on the historical data;
the output unit 202 is configured to acquire abnormal fluctuation data of a target area power grid, input the abnormal fluctuation data as risk prediction data to the deep convolution model, and identify a broadband oscillation mode of the area power grid based on the deep convolution model;
and the evaluation unit 203 is configured to collect the broadband oscillation mode data of the regional power grid identified by the deep convolution model, acquire a broadband oscillation set, and determine a broadband oscillation risk of the target regional power grid according to the broadband oscillation set.
The training unit 201 preprocesses the historical data to obtain PMU two-side-point multi-frequency amplitude data in the same time scene, including: determining abnormal fluctuation of a power grid as large fluctuation caused by non-power grid line short circuit tripping faults or unit tripping faults according to historical data, establishing a tensor for an oscillation scene at a single moment in the historical data, establishing a multi-dimensional matrix for a given power grid area, and acquiring multi-frequency amplitude data of PMU (phasor measurement unit) points at two sides in the same moment scene according to the tensor and the multi-dimensional matrix.
The training unit 201 collects historical data of regional power grid abnormal fluctuation, and obtains a deep convolution model based on the historical data, and the method comprises the following steps:
acquiring historical data of abnormal fluctuation of a regional power grid, preprocessing the historical data, acquiring multi-frequency amplitude data of two side points of a PMU (phasor measurement Unit) in the same time scene, and converting the multi-frequency amplitude data into a multi-channel color map;
and taking the multi-channel color image as input data of a pre-built deep convolution neural network, training the deep convolution neural network, and obtaining a deep convolution model.
The deep convolutional neural network pre-established in the training unit 201 comprises an activation function;
the activation function, as follows:
wherein g (x) is an activation function, e is a constant, the function value of g (x) is 0 when x <0, and the gradient is 0, and a Tanh function is output when x > 0.
The convolutional neural network pre-established in the training unit 201 further includes: a neural network architecture;
the convolution kernel of the convolutional layer of the neural network architecture is as follows:
wherein, YnFor the nth output tensor, xiIn order to input the tensor,
for the nth convolution kernel on the ith channel, bnIs the offset, and M is a constant of the total number of channels.
The training unit 201 trains the deep convolutional neural network, and includes: adjusting dynamic hyper-parameters of the deep convolutional neural network;
the dynamic hyper-parameters comprise: epoch and batch parameters, loss functions, and learning rates.
When the deep convolutional neural network is trained, the training unit 201 searches an optimal solution in a multi-process multi-start parallel search interaction manner, including:
searching an optimal solution by using N +1 CPUs, controlling each CPU to start iterative computation from an initial solution with the N being greater than 10, obtaining a computation result, and sending an adaptive value of the computation result to a unified server to be ordered according to fitness;
after the sorting is finished, selecting an operator and sorting according to the adaptive value by adopting a selection algorithm, and if the calculation result of the (N + 1) th CPU is not superior to the calculation results of the N CPUs, randomly allocating the most worried solution of the iterative calculation to the N CPUs;
and controlling the N +1 CPUs to continuously execute iterative computation, and stopping the iterative computation when the loss value is lower than the threshold value.
The output unit 202 collects the broadband oscillation mode data by using a full electromagnetic simulation network.
When the output unit 202 identifies the broadband oscillation mode of the regional power grid through the deep convolution model, if the identification fails or the identification degree does not meet the requirement, the oscillation source is locked, the broadband oscillation mode is determined according to the oscillation source, and the determined broadband oscillation mode is loaded into the stack of the deep convolution model as a new mode.
Example 3:
based on the same inventive concept, the present invention also provides a computer device comprising a processor and a memory, the memory being configured to store a computer program comprising program instructions, the processor being configured to execute the program instructions stored by the computer storage medium. The Processor may be a Central Processing Unit (CPU), or may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable gate array (FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., which is a computing core and a control core of the terminal, and is specifically adapted to implement one or more instructions, and to specifically load and execute one or more instructions in a computer storage medium so as to implement a corresponding method flow or a corresponding function, so as to implement the method steps for evaluating the risk of broadband oscillation of the regional power grid in the foregoing embodiments.
Example 4:
based on the same inventive concept, the present invention further provides a storage medium, in particular, a computer-readable storage medium (Memory), which is a Memory device in a computer device and is used for storing programs and data. It is understood that the computer readable storage medium herein can include both built-in storage medium in the computer device and, of course, extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also, the memory space stores one or more instructions, which may be one or more computer programs (including program code), adapted to be loaded and executed by the processor. It should be noted that the computer-readable storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory. One or more instructions stored in the computer-readable storage medium may be loaded and executed by a processor to implement the steps of the method for evaluating the risk of broadband oscillation of the regional power grid in the above embodiments.
The invention provides an assessment method and system for regional power grid broadband oscillation risk.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The scheme in the embodiment of the application can be implemented by adopting various computer languages, such as object-oriented programming language Java and transliterated scripting language JavaScript.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While the preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the present application.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.
Claims (20)
1. An assessment method for broadband oscillation risk of a regional power grid, the method comprising:
acquiring historical data of abnormal fluctuation of a regional power grid, and acquiring a deep convolution model based on the historical data;
acquiring abnormal fluctuation data of a target area power grid, inputting the abnormal fluctuation data serving as risk prediction data into the deep convolution model, and identifying a broadband oscillation mode of the area power grid based on the deep convolution model;
collecting the broadband oscillation mode data of the regional power grid identified by the deep convolution model to obtain a broadband oscillation set, and determining the broadband oscillation risk of the target regional power grid according to the broadband oscillation set.
2. The method according to claim 1, wherein the collecting historical data of regional power grid abnormal fluctuation, and the obtaining of the deep convolution model based on the historical data comprises:
acquiring historical data of regional power grid abnormal fluctuation, preprocessing the historical data, acquiring multi-frequency amplitude data of two side points of a PMU (phasor measurement Unit) in the same moment scene, and converting the multi-frequency amplitude data into a multi-channel color chart;
and taking the multi-channel color image as input data of a pre-built deep convolution neural network, training the deep convolution neural network, and obtaining a deep convolution model.
3. The method according to claim 2, wherein the preprocessing the historical data to obtain PMU two-side point multi-frequency amplitude data in the same time scenario comprises: determining abnormal fluctuation of a power grid as large fluctuation caused by non-power grid line short circuit tripping faults or unit tripping faults according to historical data, establishing a tensor for an oscillation scene at a single moment in the historical data, establishing a multi-dimensional matrix for a given power grid area, and acquiring multi-frequency amplitude data of PMU (phasor measurement unit) points at two sides in the same moment scene according to the tensor and the multi-dimensional matrix.
4. The method of claim 2, wherein the pre-built deep convolutional neural network comprises an activation function;
the activation function, as follows:
wherein g (x) is an activation function, e is a constant, the function value of g (x) is 0 when x <0, and the gradient is 0, and a Tanh function is output when x > 0.
5. The method of claim 2, wherein the pre-built deep convolutional neural network comprises: a neural network architecture;
the convolution kernel of the convolutional layer of the neural network architecture is as follows:
wherein, YnFor the nth output tensor, xiIn order to input the tensor,
for the nth convolution kernel on the ith channel, bnIs the offset, and M is a constant of the total number of channels.
6. The method of claim 2, wherein training the deep convolutional neural network comprises: adjusting dynamic hyper-parameters of the deep convolutional neural network;
the dynamic hyper-parameters comprise: epoch and batch parameters, loss functions, and learning rates.
7. The method of claim 2, wherein when the deep convolutional neural network is trained, searching for an optimal solution in a multi-process multi-starting point parallel search interactive manner comprises:
searching an optimal solution by using N +1 CPUs, controlling each CPU to start iterative computation from an initial solution with the N being greater than 10, obtaining a computation result, and sending an adaptive value of the computation result to a unified server to be ordered according to fitness;
after the sorting is finished, selecting an operator and sorting according to the adaptive value by adopting a selection algorithm, and if the calculation result of the (N + 1) th CPU is not superior to the calculation results of the N CPUs, randomly allocating the most worried solution of the iterative calculation to the N CPUs;
and controlling the N +1 CPUs to continuously execute iterative computation, and stopping the iterative computation when the loss value is lower than the threshold value.
8. The method of claim 1, wherein the wideband oscillation mode data is collected using a full electromagnetic simulation network.
9. The method according to claim 1, wherein when the deep convolution model identifies the broadband oscillation mode of the regional power grid, if the identification fails or the identification degree does not meet the requirement, the deep convolution model locks the oscillation source, determines the broadband oscillation mode according to the oscillation source, and loads the determined broadband oscillation mode as a new mode into a stack of the deep convolution model.
10. An assessment system for risk of wide-band oscillation of a regional power grid, the system comprising:
the training unit is used for acquiring historical data of regional power grid abnormal fluctuation and acquiring a deep convolution model based on the historical data;
the output unit is used for acquiring abnormal fluctuation data of a target regional power grid, inputting the abnormal fluctuation data serving as risk prediction data into the deep convolution model, and identifying a broadband oscillation mode of the regional power grid based on the deep convolution model;
and the evaluation unit is used for collecting the broadband oscillation modal data of the regional power grid identified by the deep convolution model, acquiring a broadband oscillation set and determining the broadband oscillation risk of the target regional power grid according to the broadband oscillation set.
11. The system of claim 10, wherein the training unit collects historical data of regional power grid abnormal fluctuation, and obtains a deep convolution model based on the historical data, and comprises:
acquiring historical data of abnormal fluctuation of a regional power grid, preprocessing the historical data, acquiring multi-frequency amplitude data of two side points of a PMU (phasor measurement Unit) in the same time scene, and converting the multi-frequency amplitude data into a multi-channel color map;
and taking the multi-channel color image as input data of a pre-built deep convolution neural network, training the deep convolution neural network, and obtaining a deep convolution model.
12. The system of claim 11, wherein the training unit preprocesses the historical data to obtain PMU dual-side point multi-frequency amplitude data in the same time scenario, and comprises: determining abnormal fluctuation of a power grid as large fluctuation caused by non-power grid line short circuit tripping faults or unit tripping faults according to historical data, establishing a tensor for an oscillation scene at a single moment in the historical data, establishing a multi-dimensional matrix for a given power grid area, and acquiring multi-frequency amplitude data of PMU (phasor measurement unit) points at two sides in the same moment scene according to the tensor and the multi-dimensional matrix.
13. The system of claim 11, wherein the deep convolutional neural network pre-built in the training unit comprises an activation function;
the activation function, as follows:
wherein g (x) is an activation function, e is a constant, the function value of g (x) is 0 when x <0, and the gradient is 0, and a Tanh function is output when x > 0.
14. The system of claim 11, wherein the deep convolutional neural network pre-constructed in the training unit comprises: a neural network architecture;
the convolution kernel of the convolutional layer of the neural network architecture is as follows:
wherein, YnFor the nth output tensor, xiIn order to input the tensor,
for the nth convolution kernel on the ith channel, bnIs the offset, and M is a constant of the total number of channels.
15. The system of claim 11, wherein the training unit trains the deep convolutional neural network, comprising: adjusting dynamic hyper-parameters of the deep convolutional neural network;
the dynamic hyper-parameters comprise: epoch and batch parameters, loss function, and learning rate.
16. The system of claim 11, wherein the training unit searches for the optimal solution in a multi-process multi-start parallel search interactive manner when training the deep convolutional neural network, and comprises:
searching an optimal solution by using N +1 CPUs, controlling each CPU to start iterative computation from an initial solution with N being greater than 10, obtaining a computation result, and sending an adaptive value of the computation result to a unified server to sequence according to fitness;
after the sorting is finished, selecting an operator and sorting according to the adaptive value by adopting a selection algorithm, and if the calculation result of the (N + 1) th CPU is not superior to the calculation results of the N CPUs, randomly allocating the most worried solution of the iterative calculation to the N CPUs;
and controlling the N +1 CPUs to continuously execute iterative computation, and stopping the iterative computation when the loss value is lower than the threshold value.
17. The system according to claim 10, wherein the output unit collects the broadband oscillation mode data using a full electromagnetic simulation network.
18. The system of claim 10, wherein when the output unit identifies the broadband oscillation mode of the local power grid through the deep convolution model, if the identification fails or the identification degree does not meet the requirement, the output unit locks the oscillation source, determines the broadband oscillation mode according to the oscillation source, and loads the determined broadband oscillation mode as a new mode into the stack of the deep convolution model.
19. A computer device, comprising:
one or more processors;
a processor for executing one or more programs;
the one or more programs, when executed by the one or more processors, implement a method for assessing risk of broadband oscillation of a regional power grid as recited in any of claims 1-9.
20. A computer-readable storage medium, having a computer program stored thereon, wherein the computer program, when executed, implements a method for assessing risk of broadband oscillation of a regional power grid according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210903376.8A CN115275990A (en) | 2022-07-27 | 2022-07-27 | Evaluation method and system for broadband oscillation risk of regional power grid |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210903376.8A CN115275990A (en) | 2022-07-27 | 2022-07-27 | Evaluation method and system for broadband oscillation risk of regional power grid |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115275990A true CN115275990A (en) | 2022-11-01 |
Family
ID=83771894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210903376.8A Pending CN115275990A (en) | 2022-07-27 | 2022-07-27 | Evaluation method and system for broadband oscillation risk of regional power grid |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115275990A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118194239A (en) * | 2024-05-15 | 2024-06-14 | 合肥工业大学 | Multi-modal graph network-based broadband oscillation parameter identification method for power transmission system |
-
2022
- 2022-07-27 CN CN202210903376.8A patent/CN115275990A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118194239A (en) * | 2024-05-15 | 2024-06-14 | 合肥工业大学 | Multi-modal graph network-based broadband oscillation parameter identification method for power transmission system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xuan et al. | Multi-model fusion short-term load forecasting based on random forest feature selection and hybrid neural network | |
| CN112383052B (en) | Power grid fault repairing method and device based on power internet of things | |
| CN112508442B (en) | Transient stability assessment method and system based on automatic and interpretable machine learning | |
| CN113177357B (en) | Transient stability assessment method for power system | |
| CN110969290A (en) | Runoff probability prediction method and system based on deep learning | |
| CN111628494B (en) | Low-voltage distribution network topology identification method and system based on logistic regression method | |
| CN113901977A (en) | Deep learning-based power consumer electricity stealing identification method and system | |
| Zheng et al. | Real-time transient stability assessment based on deep recurrent neural network | |
| CN111525587A (en) | Reactive load situation-based power grid reactive voltage control method and system | |
| CN117421571A (en) | Topology real-time identification method and system based on power distribution network | |
| CN114021483A (en) | Ultra-short-term wind power prediction method based on time domain characteristics and XGboost | |
| CN115275990A (en) | Evaluation method and system for broadband oscillation risk of regional power grid | |
| Manoj et al. | FWS-DL: forecasting wind speed based on deep learning algorithms | |
| Xu et al. | An improved ELM-WOA–based fault diagnosis for electric power | |
| CN113627674A (en) | Distributed photovoltaic power station output prediction method and device and storage medium | |
| CN111628531B (en) | Data driving method for static voltage stability evaluation of power system | |
| Zhu et al. | Prediction of air quality index based on wavelet transform combination model | |
| He et al. | Partial discharge pattern recognition algorithm based on sparse self-coding and extreme learning machine | |
| Gao et al. | Fault detection of electric vehicle charging piles based on extreme learning machine algorithm | |
| CN114415056A (en) | Distributed power supply island fault detection method | |
| Wu et al. | Convolutional deep leaning-based distribution system topology identification with renewables | |
| CN112163613A (en) | Rapid identification method for power quality disturbance | |
| Kim et al. | Cyberattack Detection in Smart Grids based on Reservoir Computing | |
| Fan et al. | Construction of a neural network and its application on target classification | |
| CN113054653A (en) | Power system transient stability evaluation method based on VGGNet-SVM |
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 |