CN116520074A - Active power distribution network fault positioning method and system based on cloud edge cooperation - Google Patents
Active power distribution network fault positioning method and system based on cloud edge cooperation Download PDFInfo
- Publication number
- CN116520074A CN116520074A CN202310307907.1A CN202310307907A CN116520074A CN 116520074 A CN116520074 A CN 116520074A CN 202310307907 A CN202310307907 A CN 202310307907A CN 116520074 A CN116520074 A CN 116520074A
- Authority
- CN
- China
- Prior art keywords
- fault
- cloud
- distribution network
- power distribution
- edge
- 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
- 238000009826 distribution Methods 0.000 title claims abstract description 101
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000012545 processing Methods 0.000 claims abstract description 27
- 230000008569 process Effects 0.000 claims description 27
- 238000011144 upstream manufacturing Methods 0.000 claims description 19
- 239000013598 vector Substances 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 14
- 238000001514 detection method Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000000007 visual effect Effects 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims 1
- 238000003745 diagnosis Methods 0.000 abstract description 13
- 238000012549 training Methods 0.000 description 39
- 230000006870 function Effects 0.000 description 26
- 238000005516 engineering process Methods 0.000 description 12
- 238000004891 communication Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 5
- 238000007726 management method Methods 0.000 description 5
- 238000003860 storage Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 238000007405 data analysis Methods 0.000 description 3
- 238000013500 data storage Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000011065 in-situ storage Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000007257 malfunction Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000002250 progressing effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 238000003892 spreading Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/088—Aspects of digital computing
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
- G01R31/086—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution networks, i.e. with interconnected conductors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
-
- 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
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Locating Faults (AREA)
Abstract
The invention discloses a cloud-edge cooperation-based active power distribution network fault positioning method and system, wherein edge computing equipment acquires fault current signals acquired by end equipment, performs ST processing on the fault current signals, and calculates total harmonic distortion rate; the edge computing equipment computes the weight root mean square value of the ST coefficient and inputs the weight root mean square value to a DBN network pre-trained by the cloud platform end; the DBN network optimizes the input weight root mean square value based on the fault characteristic attribute model according to the attribute classification factors; performing fault positioning according to the optimized weight root mean square value; according to the invention, the DBN algorithm of the cloud platform end is utilized to calculate the characteristic attribute models of different faults, and the weight root mean square value of the ST coefficient is adjusted according to the attribute classification factors so as to adapt to the fault characteristic processing and positioning of different types, and the accuracy of fault diagnosis and positioning can be improved.
Description
Technical Field
The invention relates to the technical field of protection control of grid-connected technology of a power distribution network, in particular to a cloud-edge cooperation-based active power distribution network fault positioning method and system.
Background
The cloud computing has the main advantages of mass storage, high-efficiency computing and wide-area coverage; the main advantages of the edge computing are real-time data processing, an edge resource pool, cloud computing and an application scene of the edge computing are different, and the cloud computing is good at large-scale, low in real-time requirement and long in period of big data processing and analysis; and the edge calculation is more suitable for data processing and analysis with small scale, high real-time requirement and short period. Therefore, the relationship between cloud computing and edge computing is not replaced by complementary coordination, and the requirements of various application scenes can be better met through the close coordination of the cloud and the edge, so that the application value of the cloud computing and the edge computing is enlarged. The edge calculation is close to the terminal equipment and is a data acquisition and preprocessing unit, so that cloud application can be supported; the cloud computing has strong processing capacity, can perform big data analysis optimization and model training, then issues a trained model or business rule to the edge side, and the edge computing is operated based on a new model or business rule.
The cloud edge collaboration capability and connotation mainly comprise three types of resource collaboration, data collaboration and service collaboration: 1) Resource collaboration: the edge node can provide infrastructure resources such as calculation, storage, network and the like, can independently schedule and manage local resources, can also cooperate with the cloud, and can accept and execute resource scheduling and management strategies issued by the cloud. 2) Data collaboration: the edge node is responsible for data acquisition, preprocessing and simple analysis are carried out on the original data according to a model or a business rule, and then the result and related data are uploaded to a cloud; the cloud can store, analyze and value mine mass data. The data collaboration between the edge and the cloud enables the data to flow orderly between the edge and the cloud, so that a complete data flow path is formed, and life cycle management and value mining of the data are facilitated. 3) Service collaboration: after the cloud finishes training of the model, the model is issued to the edge nodes, and the edge nodes perform reasoning according to the model; the cloud manages the life cycle of the edge side application, including deployment, starting, stopping, deleting, version updating and the like of the application; and the cloud end generates an application arrangement strategy, and the edge side executes the application according to the cloud end strategy.
Cloud edge coordination has three modes (1) training-cloud edge coordination of calculation. The cloud end designs, trains and updates the intelligent model according to the data uploaded by the edge, and the edge end is responsible for collecting the data and downloading the latest model in real time to carry out calculation tasks; and (2) cloud-oriented cloud edge collaboration. The cloud end can bear the design, training and updating of the intelligent model, also can bear the calculation task of the front section of the model, and then transmits the intermediate result to the edge end, so that the edge end can continue to calculate to obtain a final result. The mode aims at weighing the calculated amount and the traffic of the cloud end and the edge end; and (3) edge-guided cloud edge cooperation. The cloud is only responsible for initial training work, and the model training is downloaded to the edge after completion. The edge performs the computing task while also utilizing the real-time in-situ data to perform subsequent training of the model. This mode aims at meeting the personalized needs of the application, making better use of the local data.
The power distribution network fault diagnosis judges whether the system has faults or not by monitoring voltage, current, event information and the like in the running process of the system, and identifies the type and reason of the faults; the fault location of the distribution network is to use the measured voltage, current or other parameter information to timely and reliably give out the fault area and the accurate position. The traditional power distribution network is mainly a radial network of a single power supply, and the tide has unidirectionality, namely, the current flows to the load from a bus of a transformer substation, and the fault direction does not need to be judged. In order to simplify the protection configuration and reduce the construction cost of the power distribution network, the conventional single-ended three-section current protection, namely current quick-break protection, time-limited current quick-break protection and time-limited overcurrent protection, is commonly configured for the power distribution network at present, and the functions of fault diagnosis and positioning are realized through the coordination of fixed values and time limits between different protection elements at the upstream and downstream of the feeder line.
However, when DG is present in the distribution network, the energy flow changes the original unidirectional flow restriction, the coordinated operation of the conventional fault protection scheme and the application of feeder automation are affected by multiple power sources, and the fault diagnosis and localization mainly have the following problems: the fault characteristic extraction and analysis are difficult: DG in active distribution networks are typically connected to the distribution network via power electronics. The fault current is limited due to the fact that a large number of power electronic devices are connected into an active power distribution network, fault characteristics are not obvious, harmonic waves are complex, and the accuracy of fault diagnosis and positioning is seriously affected. (2) the traditional fault diagnosis and positioning method is difficult to be applied to: the distributed access of DG makes the distribution network change from traditional passive distribution network to more active distribution network, its topological structure is complicated, the operation mode is nimble, characteristics such as the bidirectional changeable of trend and output uncertainty lead to distribution network fault feature to have great difference with traditional distribution network, distribution network trouble and electric energy quality problem intercoupling, distribution network trouble presents weak characteristic and high frequency transient state characteristic more obvious, distribution network fault diagnosis and processing degree of difficulty increase, make the current relay protection of distribution network, fault diagnosis and locate performance reduce, take place malfunction and misjudgement easily, seriously threat distribution network safe operation.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the method and the system for locating the faults of the active power distribution network based on cloud edge cooperation can improve the accuracy of fault diagnosis and locating.
In order to solve the technical problems, the invention adopts the following technical scheme:
a cloud-edge cooperation-based active power distribution network fault positioning method comprises the following steps:
s1, an edge computing device acquires a fault current signal acquired by an end device, performs ST processing on the fault current signal, and calculates a total harmonic distortion rate;
s2, calculating a weight root mean square value of the ST coefficient by edge calculation equipment, and inputting the weight root mean square value to a DBN network pre-trained by a cloud platform end;
s3, the DBN network optimizes the input weight root mean square value based on the fault characteristic attribute model according to the attribute classification factors;
and S4, performing fault positioning according to the optimized weight root mean square value.
In order to solve the technical problems, the invention adopts another technical scheme that:
the active power distribution network fault positioning system based on cloud edge coordination comprises a cloud platform end, edge computing equipment and end equipment, wherein the cloud platform end, the edge computing equipment and the end equipment jointly realize the steps in the active power distribution network fault positioning method based on cloud edge coordination.
The invention has the beneficial effects that: according to the fault positioning method and system for the active power distribution network based on cloud edge cooperation, firstly, fault current signals are captured at edge computing equipment based on ST processing, on-line fault line detection is carried out on the lines by using total harmonic distortion rate THD of fault current signals at the end parts of the lines, a weight root mean square value RMS is calculated according to ST coefficients of obtained detected signals, fault feeder lines and lines are detected through THD levels of the fault signals, finally, characteristic attribute models of different faults are obtained by using DBN algorithm of a cloud platform end, the weight root mean square value of the ST coefficients is adjusted according to attribute classification factors, so that the method is suitable for fault characteristic processing and positioning of different types, and fault diagnosis and positioning accuracy can be improved.
Drawings
FIG. 1 is a schematic diagram of a DBN network based on a cloud-edge cooperative active power distribution network fault location method according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a power distribution internet of things architecture of an active power distribution network fault location system based on cloud-edge coordination according to an embodiment of the present invention;
FIG. 3 is a flowchart of an active power distribution network fault location method based on cloud edge coordination according to an embodiment of the present invention;
fig. 4 is a main flowchart of an active power distribution network fault positioning method based on cloud edge coordination according to an embodiment of the invention;
fig. 5 is a schematic diagram of a layer-by-layer pre-training process of a DBN of an active power distribution network fault location method based on cloud edge coordination according to an embodiment of the present invention;
fig. 6 is a schematic diagram of a DBN fine tuning process of an active power distribution network fault location method based on cloud-edge coordination according to an embodiment of the present invention.
Detailed Description
In order to describe the technical contents, the achieved objects and effects of the present invention in detail, the following description will be made with reference to the embodiments in conjunction with the accompanying drawings.
ST: stockwell transform, hyperbolic Stockwell transform;
THD: total Harmonic Distortion, total harmonic distortion.
Referring to fig. 1, 3 and 4, a cloud-edge collaboration-based active power distribution network fault positioning method includes the steps of:
s1, an edge computing device acquires a fault current signal acquired by an end device, performs ST processing on the fault current signal, and calculates a total harmonic distortion rate;
s2, calculating a weight root mean square value of the ST coefficient by edge calculation equipment, and inputting the weight root mean square value to a DBN network pre-trained by a cloud platform end;
s3, the DBN network optimizes the input weight root mean square value based on the fault characteristic attribute model according to the attribute classification factors;
and S4, performing fault positioning according to the optimized weight root mean square value.
From the above description, the beneficial effects of the invention are as follows: according to the fault positioning method and system for the active power distribution network based on cloud edge cooperation, firstly, fault current signals are captured at edge computing equipment based on ST processing, on-line fault line detection is carried out on the lines by using total harmonic distortion rate THD of fault current signals at the end parts of the lines, a weight root mean square value RMS is calculated according to ST coefficients of obtained detected signals, fault feeder lines and lines are detected through THD levels of the fault signals, finally, characteristic attribute models of different faults are obtained by using DBN algorithm of a cloud platform end, the weight root mean square value of the ST coefficients is adjusted according to attribute classification factors, so that the method is suitable for fault characteristic processing and positioning of different types, and fault diagnosis and positioning accuracy can be improved.
Further, the step S1 further includes the following steps:
s1, performing fault detection by the terminal equipment, and acquiring a fault current signal.
Further, the coding rule of the fault current signal is:
the switching function is:
wherein,,the switch j is composed of two parts, and the switch j divides an upstream switch and a downstream switch; />And->Respectively an upstream switch function and a downstream switch function; m and N are the total number of upstream and downstream segments, respectively; m is M 1 And N 1 The total number of the upper power supply and the lower power supply is respectively; k (K) u And K d The power supply access conditions of the upstream and the downstream are respectively, wherein a value of 1 indicates access and a value of 0 indicates unconnected; />And->Node states of the switch j to the upstream and downstream line power supply sections respectively; s is(s) j,u Sum s j,d The node states of the upstream section and the downstream section of the switch j are respectively, wherein a value of 1 indicates a fault, and a value of 0 indicates a normal.
As can be seen from the above description, after the distributed power generation device DG is connected to the grid, the DG responds to different fault conditions, and the fault current direction is different. At this time, the use of only 0 and 1 does not represent well the fault current condition flowing through the node. Therefore, the invention introduces a new coding rule, and considers that the fault current flowing through the switch on the branch of the power distribution network connected with the DG may be opposite to the fault current flowing in the direction before the DG is connected with the branch, and the fault current state of the switch uses a code of "-1". And after DG is connected, the power flow of the power distribution network changes along with the switching of DG, and the current direction of each node also changes. In the event of a fault, the current direction monitored by each node is not unique, but is determined by the current information of each power supply, so the invention also makes a new functional representation of the switching function.
Further, in the fault detection process of the terminal equipment, the adopted fitness function is as follows:
where ω represents a weight factor.
As can be seen from the above description, in the process of locating the fault section of the power distribution network, that is, the process of minimizing the difference between the fault information uploaded by the end device (feeder terminal unit FTU) and the expected value derived by the switching function, in order to prevent erroneous judgment, the fitness function as described above is introduced.
Further, the fault current signal is expressed as:
x(t)＝[x(1),x(2),…,x(m)]；
wherein x (m) represents a fault acquisition signal of the mth terminal device;
the ST process is specifically as follows:
where t is the time parameter, τ is the position of the control window function on the time axis, f is the frequency,is a window function;
after ST processing is carried out on the fault current signal, a two-dimensional matrix A with time and frequency on the abscissa is obtained:
wherein F is ST treatment; t is t k f j For the magnitude of the two-dimensional matrix A, k ε (1, m), j ε (1, n).
As is clear from the above description, based on the above steps, ST processing is performed on the fault current signal.
Further, the calculation of the weighted root mean square value is specifically as follows:
wherein F is RMS Weighted root mean square value, N, representing ST coefficient s Representing the harmonic quantity, σ represents the width of the gaussian function, and THD represents the total harmonic distortion.
From the above description, the weighted root mean square value of the ST coefficient is calculated as above.
Further, the calculation of the total harmonic distortion rate specifically includes:
wherein Q represents the total effective value of the current, Q 1 Representing the line fundamental effective value.
From the above description, the total harmonic distortion ratio THD is calculated as above.
Further, the DBN network is considered as a stack of boltzmann machines RBMs;
here, the RBM state is represented as (v, h), and the energy provided is represented as:
wherein v is i A fault state vector being a visual layer element; h is a j A fault state vector for the hidden layer unit; θ is the RBM parameter set; w (W) ij The connection weight value of the visible layer unit and the hidden unit is used; a, a i 、b j The bias vectors of the visual element and the hidden element, respectively.
From the above description, it is clear that the DBN network can be referred to as above.
Referring to the figure, the active power distribution network fault positioning system based on cloud-edge coordination comprises a cloud platform end, edge computing equipment and end equipment, wherein the cloud platform end, the edge computing equipment and the end equipment jointly realize the steps in the active power distribution network fault positioning method based on cloud-edge coordination.
The method and the system for locating the faults of the active power distribution network based on cloud edge cooperation are suitable for fault diagnosis and locating of the active power distribution network.
Referring to fig. 1, 3 and 4, a first embodiment of the present invention is as follows:
a cloud-edge cooperation-based active power distribution network fault positioning method comprises the following steps:
s0, the terminal equipment performs fault detection and acquires fault current signals.
In this embodiment, when a short-circuit fault occurs in a certain line in the power distribution network, the Feeder Terminal Unit (FTU) uploads the detection information to the control center. And finally finding out a fault process through optimizing an integrated intelligent algorithm.
si is the state of a certain part of the power distribution network, and the specific expression method is shown as the formula (1):
DG is not connected with the network, only two line states including fault and normal are adopted in fault positioning, and FTU collected data of the corresponding switch are only normal and overcurrent. Thus, FTU acquisition information may be encoded as shown in equation (2):
after grid connection of DGs (Distributed Generation, distributed power generation devices), the DGs respond to different fault conditions and the fault current directions are different. At this time, the use of only 0 and 1 does not represent well the fault current condition flowing through the node. Therefore, a new coding rule is introduced as shown in formula (3):
considering that the fault current flowing through a switch on a branch of a distribution network connected to a DG may be opposite to the fault current flowing before the DG is connected, the fault current state of the switch is used with a code of minus 1'.
Building a switch model:
before DG is disconnected, the switching function is as shown in formula (4):
wherein I is j (s) is the switching function of the j-th switch; s is(s) i A state of the downstream section i; pi is a logical OR.
After DG is connected, the power flow of the power distribution network changes along with the switching of DG, and the current direction of each node also changes. Furthermore, in the event of a fault, the current direction monitored by each node is not unique, but is determined by the current information of each power supply.
The switching function used in this embodiment is shown in equation (5):
wherein,,the switch j is composed of two parts, and the switch j divides an upstream switch and a downstream switch; />And->Respectively an upstream switch function and a downstream switch function; m and N are the total number of upstream and downstream segments, respectively; m is M 1 And N 1 The total number of the upper power supply and the lower power supply is respectively; k (K) u And K d The power supply access conditions of the upstream and the downstream are respectively, wherein a value of 1 indicates access and a value of 0 indicates unconnected; />And->Node states of the switch j to the upstream and downstream line power supply sections respectively; s is(s) j,u Sum s j,d The node states of the upstream section and the downstream section of the switch j are respectively, wherein a value of 1 indicates a fault, and a value of 0 indicates a normal.
Constructing a fitness function:
in order to prevent erroneous judgment, the following fitness function is adopted in the embodiment:
based on the architecture and the fault positioning model, the method provided by the embodiment embeds ST processing into the internet-of-things proxy terminal, and utilizes the computing capability of the edge side (edge computing equipment) to detect and preprocess the fault information in situ; and then different fault characteristic training, learning classification and fault positioning are carried out by using the DBN through a platform of a cloud master station (cloud platform end).
S1, an edge computing device acquires a fault current signal acquired by an end device, ST processes the fault current signal, and calculates the total harmonic distortion rate.
Feature extraction at edge side based on ST processing:
fault current signals collected by an intelligent terminal (terminal equipment) at the terminal side:
x(t)＝[x(1),x(2),…,x(m)]；
wherein x (m) represents a fault acquisition signal representing an mth terminal device;
the ST process in the edge side is specifically:
where t is the time parameter, τ is the position of the control window function on the time axis, f is the frequency,is a window function;
after ST processing is carried out on the fault current signal, a two-dimensional matrix A with time and frequency on the abscissa is obtained:
wherein F is ST treatment; t is t k f j For the magnitude of the two-dimensional matrix A, k ε (1, m), j ε (1, n).
THD changes according to the distributed power supply access amount, if the effective value of the line fundamental wave is Q1 and the total effective value of the current is Q, the total harmonic distortion THD of the current signal is expressed as:
s2, the edge computing equipment computes a weight root mean square value of the ST coefficient and inputs the weight root mean square value to the DBN network pre-trained by the cloud platform end.
In this embodiment, root Mean Square (RMS) of ST coefficients takes into account the varying influence of THD when distributed power is accessed, which varies from one fault location to another. The ST coefficient weight root mean square value is selected from the extracted characteristics and used as an input parameter of cloud side (cloud platform end) DBN training, and the expression is as follows:
wherein F is RMS Weighted root mean square value, N, representing ST coefficient s Representing the harmonic quantity, σ represents the width of the gaussian function, and THD represents the total harmonic distortion.
And S3, the DBN network optimizes the input weight root mean square value based on the fault characteristic attribute model according to the attribute classification factors.
Feature attribute model: is a mathematical model that includes signals of the acquisition faults, which is the primary basis for identifying faults.
Attribute classification factor: the collected signals are classified through factor coefficients so as to distinguish fault characteristics, and the attribute classification is optimized so that the fault identification rate is higher and more accurate.
And S4, performing fault positioning according to the optimized weight root mean square value.
In this embodiment, the DBN learning process is adopted to perform fault feature training, perform attribute classification on the fault, and then reversely optimize the weighted root mean square values of ST coefficients of different types of faults.
The DBN is used as an unsupervised deep learning neural network to actively learn input data and automatically mine rich information hidden in known data. The model structure is shown in fig. 1.
The DBN can be regarded as a stack of boltzmann machines (restricted boltzmann machines, RBM), and the RBM states are expressed as (v, h), and the energy provided is expressed as:
wherein v is i A fault state vector being a visual layer element; h is a j A fault state vector for the hidden layer unit; θ is the RBM parameter set; w (W) ij The connection weight value of the visible layer unit and the hidden unit is used; a, a i 、b j The bias vectors of the visual element and the hidden element, respectively.
The RBM learning model aims to solve characteristic attribute models of different faults, and the weight root mean square value of the ST coefficient is adjusted according to the attribute classification factors so as to adapt to different types of fault characteristic processing.
The ST-DBN based power grid fault locating flow is shown in fig. 3. Fig. 3 (a) is a fault line detection flow based on THD and ST, and fig. 3 (b) is a final fault location process including a measurement module, a signal processing module, and a DBN training learning module.
Fig. 3 (a) is only a simple illustration of the process, and the present invention implements the final fault localization process.
Firstly, based on ST processing at the edge side, a fault current signal is captured from an intelligent terminal at the end side, and online fault line detection is carried out on the line by utilizing the THD value of the fault current signal at the end part of the line. And then, calculating an RMS value according to the ST coefficient of the obtained measured signal, and detecting the fault feeder line and the line through the THD level of the fault signal. And finally, calculating characteristic attribute models of different faults by using a DBN algorithm at the cloud side, and adjusting the weighted root mean square value of the ST coefficient according to the attribute classification factors so as to adapt to the processing and positioning of different types of fault characteristics.
The second embodiment of the invention is as follows:
the difference between the cloud edge cooperation-based active power distribution network fault positioning method and the first embodiment is that the training process of the DBN is described in the first embodiment.
The DBN may be regarded as a stack of boltzmann machines (restricted boltzmann machines, RBM), in this embodiment, the DBN adopts a layer-by-layer training method, only one layer of limited boltzmann machine (Restricted Boltzmann Machine, RBM) RBM network is trained at a time, parameters of each layer of RBM network are respectively adjusted, output data of the former layer is used as input data of the latter layer, and after all training of the RBM network is completed, a BP neural network algorithm is then used to perform reverse fine tuning in combination with tag values of the data. The Training process of the DBN prediction model is divided into two stages of Pre-Training (Pre-Training) and Fine-Tuning (Fine-Tuning).
1. Layer-by-layer pre-training
The first stage is a greedy pre-training learning process of stacking RBMs, and in order to optimize parameters such as bias, weight and the like of the neural network, a Contrast Divergence (CD) algorithm of unsupervised layer-by-layer learning is adopted to initialize the parameters of the neural network. And taking output parameters of the RBM of the upper layer through the unsupervised learning neural network model as input of the RBM of the lower layer until the RBM of each layer is trained, and storing corresponding weights as priori knowledge of the parameter learning process of the second stage.
In this example, the RBM was trained using a contrast divergence (Contrastive Divergence, CD) algorithm. In the CD algorithm, first use is made of
The probabilities of all hidden variables are calculated and, based on this distribution, a hidden vector is sampled from it by the gibbs sampling method. Based on the hidden vector, the utilization type is utilized
The probability of observable variables is calculated and from this distribution the input vector of the visible layer is reconstructed again by gibbs sampling. This mapping and reconstruction (reconstruction) process between the visible layer and the hidden layer is repeatedly performed, and the values of weights a, b and W are continuously updated during each reconstruction. The updating criteria of each parameter are as follows:
ΔW ij ＝e(<v i h j > data -〈v i h j 〉 recon )；
Δa i ＝e(〈v i 〉 data -〈v i 〉 rec o n )；
Δb j ＝(〈h j 〉 data -<h j > recon )；
where e is learning rate, e.e. (0, 1).<.> data Representing the mathematical expectation defined by the input dataset,<.> recon representing the mathematical expectation of model definition after one-step reconstruction.
In general, instead of waiting to converge, only k times are needed to alternately gibbs sample the guest observation vector and the hidden vector, which is the CD-k algorithm. Generally, k=1 can be trained to better parameters.
The DBN layer-by-layer pre-training process is shown in fig. 5, using training samples as input vectors for the first RBM for coefficient matrix training. That is, vectors generated by the first RBM observable layer are transferred to the hidden layer using the gibbs sampling method. Meanwhile, the observable layer is reconstructed by using the hidden layer, and the weight matrix of the network parameters is updated by using the difference between the reconstructed layer and the observable layer. Training a first RBM by adopting a CD algorithm until the set maximum training times are reached, determining the weight and bias parameters of the first RBM by the training result, acquiring the state of the hidden layer node by combining an activation function, randomly training a second RBM by adopting the CD algorithm, stacking the second RBM on the first RBM after the set maximum training times are met, namely, after the parameter training is finished, training layer by layer, and progressing layer by layer until all RBMs are trained.
2. Fine tuning
The second stage is a backward fine tuning learning process of the DBN: the network parameters are globally fine-tuned using a Back propagation algorithm (BP).
As shown in fig. 6, after each RBM layer-by-layer training is finished, adding a regression or classification layer as an output layer at the top of the obtained DBN, and then gradually transmitting to the bottom layer by using a back propagation algorithm from the top layer of the whole network, namely the final output layer, and expanding tuning operation on all hidden layers until the output value and the actual value of the network reach the range of the receiving errors of the training design or reach the set maximum iteration number to terminate. For regression problems, the top layer of the network is generally a regression layer with only one node.
In general, the reverse fine tuning process adopts BP algorithm to distribute errors to each RBM in the DBN network structure from top to bottom, thereby realizing the supervised training process of the whole network and further realizing the global further optimization of the parameters of the model. The problem that the BP algorithm is easy to fall into local optimum and leads to lengthening of training time due to random initialization of weight parameters is avoided, the convergence speed of the tuning stage is increased, and the prediction accuracy of the model is improved.
The specific algorithm flow of the reverse fine tuning in the DBN is as follows:
(1) Parameter initialization: initializing DBN parameters by using weight and bias learned by pre-training, and inputting training sample vector
(2) Training parameters: selecting corresponding activation functions, spreading input data to the top layer by layer, selecting a required regression or classification function by the top layer of the network, and training the whole network layer by layer to obtain an output value of model prediction;
(3) Calculating training errors: calculating the error between the predicted value and the actual value to obtain a gradient value of the corresponding weight and the bias parameter;
weight adjustment: and (3) reversely adjusting parameters by a gradient descent method, continuously updating weights and bias vectors until the error of the predicted value and the actual value reaches a specific value or reaches set iteration times, and ending the training network.
Referring to fig. 2, a third embodiment of the present invention is as follows:
the cloud edge cooperation-based active power distribution network fault positioning system comprises a cloud platform end, edge computing equipment and end equipment, wherein the cloud platform end, the edge computing equipment and the end equipment jointly realize the steps in the cloud edge cooperation-based active power distribution network fault positioning method in the first embodiment.
In this embodiment, a specific description is given to a cloud-edge cooperative architecture of a power distribution network based on a cloud-edge cooperative active power distribution network fault positioning system:
referring to fig. 2, the cloud-edge cooperative architecture of the power distribution network in this embodiment adopts a four-layer scheme of "cloud-pipe-edge-end".
Cloud platform end: the "cloud" platform layer, the distribution cloud master station. Besides unified asset management of equipment in the power distribution network, receiving control instructions from an upper dispatching center, accurately controlling the equipment in the power distribution network and other traditional power distribution network services, the power distribution cloud master station also bears training tasks of various deep learning models due to the introduction of advanced technologies such as cloud computing technology, distributed storage technology, big data analysis and mining technology, artificial intelligence technology and the like, the operation intellectualization of the original services is improved by utilizing an advanced deep learning method, and the novel service scene of the power distribution network is possibly expanded. Therefore, the power distribution cloud master station needs to realize unified allocation of cloud and side network resources, storage resources, computing resources and the like to the power distribution network by adopting a software definition scheme on the basis of meeting the requirements of traditional power distribution automation business, equipment asset data penetration and information fusion, plug and play of terminal equipment, real-time graph and model maintenance, and finally realizes the reliability and economy of the integral operation of the power distribution network by adopting a container technology, a virtual pool technology, a parallelization computing technology and other means, and carrying out safe, economic and refined regulation and control on traditional power resources (such as energy storage equipment, voltage regulating equipment, new energy stations and the like) of the power distribution network on the basis of cloud computing technology, artificial intelligence technology and the like. Therefore, the power distribution cloud master station can be further subdivided into the following 3 layers, and differentiated and personalized services are provided for different users and application scenes of the power distribution network:
(1) Infrastructure as a service layer: and virtualizing and pooling the computing resources of all cloud ends and side ends in the power distribution network, so that sufficient computing resources and data storage space are provided for computing tasks such as solving of optimal models of various regulation and control tasks in the power distribution network, training of models based on deep learning and the like.
(2) Platform is service layer: and a unified data interaction interface and a storage processing function are provided for various power distribution network application programs, so that an application program developer does not need to directly manage or control cloud computing resources of the power distribution network, and development efficiency is improved.
(3) Software is the service layer: through the containerization technology and the virtualization scheme, according to actual power business scenes, micro applications and micro services such as asset management, state monitoring and automatic operation and maintenance are deployed in the cloud, and intelligent energy consumption requirements of operation and maintenance of the power distribution network and users are met.
"pipe" layer: i.e. the network management layer. As the name implies, the system has the function of providing an information interaction pipeline with the service quality guarantee for various types of services (such as distribution automation, marketing metering, situation awareness, intelligent operation and maintenance and the like) of the distribution network. The specific communication modes can be divided into two types of wired communication and wireless communication according to communication media, wherein the wired communication comprises optical fiber communication, RS232, RS485 and the like; the wireless communication includes a long-distance mobile air network including a 5G and 230MHz power private network, and also includes a short-distance wireless transmission mode including wifi, zig Bee, bluetooth, power line carrier and the like.
Edge computing device: namely an intelligent fusion terminal. It serves as a distributed intelligent agent closer to the data source and on the network edge side than the distribution cloud master, with some data storage, calculation and analysis capabilities, providing autonomous decisions and services on-site or nearby. In the actual construction process of cloud-edge coordination of the power distribution network, the intelligent fusion terminal is often deployed in a power distribution station area close to a user side and an intelligent substation, perceives and locally pre-processes data in the area, and simultaneously provides computing resource support and coordination for a power distribution cloud master station, so that excessive load on a communication network is avoided when bottom mass perceiving data are uploaded to the power distribution cloud master station in a centralized mode, data storage and analysis pressure of the power distribution cloud master station are reduced, and overall computing efficiency is improved.
End device: i.e. distribution network terminal physical devices such as various voltage regulating devices, inverter devices, switching devices, as well as various types of sensors (e.g. voltage/current sensors, SF6 content monitors, etc.), feeder Terminal Units (FTUs), etc. The terminal equipment is responsible for state sensing and execution control main body in the power distribution network, the obtained mass data can provide running states of all links of the power distribution network for a power distribution cloud main station, solid data sources are provided for panoramic situation sensing of the power distribution network, and various advanced application scenes of the power distribution network are finally supported.
In summary, according to the active power distribution network fault positioning method and system based on cloud edge cooperation provided by the invention, firstly, based on ST processing, fault current signals are captured at edge computing equipment, on-line fault line detection is carried out on a line by utilizing Total Harmonic Distortion (THD) of line end fault current signals, a root mean square value (RMS) is calculated according to ST coefficients of obtained detected signals, fault feeder lines and lines are detected through the THD levels of the fault signals, finally, characteristic attribute models of different faults are obtained by utilizing DBN algorithm of a cloud platform end, and the root mean square value of the ST coefficients is adjusted according to attribute classification factors so as to adapt to different types of fault characteristic processing and positioning, and the accuracy of fault diagnosis and positioning can be improved.
The foregoing description is only illustrative of the present invention and is not intended to limit the scope of the invention, and all equivalent changes made by the specification and drawings of the present invention, or direct or indirect application in the relevant art, are included in the scope of the present invention.
Claims (9)
1. The active power distribution network fault positioning method based on cloud edge cooperation is characterized by comprising the following steps:
s1, an edge computing device acquires a fault current signal acquired by an end device, performs ST processing on the fault current signal, and calculates a total harmonic distortion rate;
s2, calculating a weight root mean square value of the ST coefficient by edge calculation equipment, and inputting the weight root mean square value to a DBN network pre-trained by a cloud platform end;
s3, the DBN network optimizes the input weight root mean square value based on the fault characteristic attribute model according to the attribute classification factors;
and S4, performing fault positioning according to the optimized weight root mean square value.
2. The method for locating faults of an active power distribution network based on cloud edge collaboration according to claim 1, wherein the step S1 is preceded by the steps of:
s0, the terminal equipment performs fault detection and acquires fault current signals.
3. The fault location method for the active power distribution network based on cloud edge cooperation according to claim 2, wherein the coding rule of the fault current signal is as follows:
the switching function is:
wherein,,the switch j is composed of two parts, and the switch j divides an upstream switch and a downstream switch; />And->Respectively an upstream switch function and a downstream switch function; m and N are the total number of upstream and downstream segments, respectively; m is M 1 And N 1 The total number of the upper power supply and the lower power supply is respectively; k (K) u And K d Respectively are provided withFor the power supply access conditions of the upstream and downstream, a value of 1 indicates access, and a value of 0 indicates unconnected; />And->Node states of the switch j to the upstream and downstream line power supply sections respectively; s is(s) j,u Sum s j,d The node states of the upstream section and the downstream section of the switch j are respectively, wherein a value of 1 indicates a fault, and a value of 0 indicates a normal.
4. The method for locating faults of the active power distribution network based on cloud edge coordination according to claim 3, wherein in the fault detection process of the terminal equipment, an adaptive function is adopted as follows:
where ω represents a weight factor.
5. The cloud edge collaboration-based active power distribution network fault location method as claimed in claim 1, wherein the fault current signal is expressed as:
x(t)＝[x(1),x(2),…,x(m)]；
wherein x (m) represents a fault acquisition signal of the mth terminal device;
the ST process is specifically as follows:
where t is the time parameter, τ is the position of the control window function on the time axis, f is the frequency,is a window function;
after ST processing is carried out on the fault current signal, a two-dimensional matrix A with time and frequency on the abscissa is obtained:
wherein F is ST treatment; t is t k f j For the magnitude of the two-dimensional matrix A, k ε (1, m), j ε (1, n).
6. The cloud edge collaboration-based active power distribution network fault location method as claimed in claim 5, wherein the calculation of the weighted root mean square value is specifically as follows:
wherein F is RMS Weighted root mean square value, N, representing ST coefficient s Representing the harmonic quantity, σ represents the width of the gaussian function, and THD represents the total harmonic distortion.
7. The cloud edge collaboration-based active power distribution network fault location method according to claim 1 or 6, wherein the calculation of the total harmonic distortion rate is specifically:
wherein Q represents the total effective value of the current, Q 1 Representing the line fundamental effective value.
8. The cloud-edge collaboration-based active power distribution network fault location method of claim 1, wherein the DBN network is considered as a stack of boltzmann machines RBMs;
here, the RBM state is represented as (v, h), and the energy provided is represented as:
wherein v is i A fault state vector being a visual layer element; h is a j A fault state vector for the hidden layer unit; θ is the RBM parameter set; w (W) ij The connection weight value of the visible layer unit and the hidden unit is used; a, a i 、b j The bias vectors of the visual element and the hidden element, respectively.
9. An active power distribution network fault location system based on cloud edge coordination comprises a cloud platform end, edge computing equipment and end equipment, wherein the cloud platform end, the edge computing equipment and the end equipment jointly realize the steps in an active power distribution network fault location method based on cloud edge coordination according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310307907.1A CN116520074A (en) | 2023-03-27 | 2023-03-27 | Active power distribution network fault positioning method and system based on cloud edge cooperation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310307907.1A CN116520074A (en) | 2023-03-27 | 2023-03-27 | Active power distribution network fault positioning method and system based on cloud edge cooperation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116520074A true CN116520074A (en) | 2023-08-01 |
Family
ID=87393003
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310307907.1A Pending CN116520074A (en) | 2023-03-27 | 2023-03-27 | Active power distribution network fault positioning method and system based on cloud edge cooperation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116520074A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454234A (en) * | 2023-12-26 | 2024-01-26 | 国网天津市电力公司宁河供电分公司 | County power grid fault identification method and county power grid fault identification device based on cloud edge cooperation |
-
2023
- 2023-03-27 CN CN202310307907.1A patent/CN116520074A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454234A (en) * | 2023-12-26 | 2024-01-26 | 国网天津市电力公司宁河供电分公司 | County power grid fault identification method and county power grid fault identification device based on cloud edge cooperation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059356A (en) | A kind of bulk power grid intelligent control system and method based on big data and artificial intelligence | |
| CN105825271B (en) | Satellite failure diagnosis and prediction method based on evidential reasoning | |
| CN116245033B (en) | Artificial intelligent driven power system analysis method and intelligent software platform | |
| CN109490814A (en) | Metering automation terminal fault diagnostic method based on deep learning and Support Vector data description | |
| Ismail et al. | A hybrid model of self organizing maps and least square support vector machine for river flow forecasting | |
| CN116520074A (en) | Active power distribution network fault positioning method and system based on cloud edge cooperation | |
| CN116914751B (en) | Intelligent power distribution control system | |
| CN117110748A (en) | Transformer substation main equipment operation state abnormality detection method based on fusion terminal | |
| CN112329973A (en) | Space-time load prediction method based on graph neural network and regional gridding | |
| Xiao et al. | Online sequential extreme learning machine algorithm for better predispatch electricity price forecasting grids | |
| Muhammad et al. | Deep-reinforcement-learning-based sustainable energy distribution for wireless communication | |
| CN115603446A (en) | Power distribution station area operation monitoring system based on convolution neural network and cloud edge synergistic effect | |
| Salgado et al. | A short-term bus load forecasting system | |
| Zhao et al. | Optimizing electric vehicle charging schedules and energy management in smart grids using an integrated GA-GRU-RL approach | |
| CN115277789B (en) | Safety protection system and method for cascade hydropower station | |
| CN116758740A (en) | Intelligent traffic efficient simulation method based on digital twin and information tree | |
| CN110827134A (en) | Power grid enterprise financial health diagnosis method | |
| CN113421004B (en) | Transmission and distribution cooperative active power distribution network distributed robust extension planning system and method | |
| CN111857015B (en) | Power transmission and transformation cloud intelligent controller | |
| Zheng et al. | [Retracted] Application Based on Artificial Intelligence in Substation Operation and Maintenance Management | |
| Simões et al. | Advances in information technology for Smart Grids | |
| Dutta et al. | Machine Learning‐Based Intelligent Power Systems | |
| Keyan et al. | Anomaly detection method of distribution network line loss based on hybrid clustering and LSTM | |
| CN111897655B (en) | Power transmission and transformation high in clouds intelligent control ware | |
| Jha et al. | Machine Learning and Deep Learning Approaches for Energy Management in Smart Grid 3.0 |
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 |