CN114915035B - Power distribution network monitoring method, device and system - Google Patents
Power distribution network monitoring method, device and system Download PDFInfo
- Publication number
- CN114915035B CN114915035B CN202210846447.5A CN202210846447A CN114915035B CN 114915035 B CN114915035 B CN 114915035B CN 202210846447 A CN202210846447 A CN 202210846447A CN 114915035 B CN114915035 B CN 114915035B
- Authority
- CN
- China
- Prior art keywords
- state
- order
- power distribution
- distribution network
- monitoring
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
- H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
- H02J13/00002—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by monitoring
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
- G16Y40/00—IoT characterised by the purpose of the information processing
- G16Y40/10—Detection; Monitoring
-
- 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
- H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
- H02J13/00001—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by the display of information or by user interaction, e.g. supervisory control and data acquisition systems [SCADA] or graphical user interfaces [GUI]
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
Abstract
The embodiment of the invention provides a monitoring method, a device and a system for a power distribution network, which relate to the field of power distribution and Internet of things, and the monitoring method comprises the following steps: acquiring state information in the power distribution network in real time, wherein the state information is at least one of a state video vector, a state image vector and a state text vector; taking the state information as particles to be brought into a PSO particle swarm algorithm to determine optimal parameters of the particles; bringing the optimal parameters of the particles into a BilSTM model to obtain a key feature set; determining attention degrees of different features in the key feature set by combining an AM algorithm and the BilSTM model; monitoring for faults in the power distribution network based on the attentiveness of the characteristic. The monitoring method improves the accuracy and convergence time of power distribution network monitoring fault positioning, realizes the flow control of the power distribution terminal equipment fault, and reduces the operation risk of the terminal equipment.
Description
Technical Field
The invention relates to the field of power distribution and Internet of things, in particular to a method, a device and a system for monitoring a power distribution network.
Background
With the development of the power distribution internet of things, the platform area intelligent terminal is used as an edge computing node of the low-voltage power distribution internet of things and is mainly deployed near a low-voltage power distribution transformer, core capabilities of network, computing, storage and application are integrated, edge intelligent service is provided nearby, and value flow of low-voltage platform area data, modeling of heterogeneous data and business application of the heterogeneous data and cooperation of all links of a cloud edge end are achieved.
Towards lean fortune and comprehensive energy management, novel fortune dimension devices such as camera, intelligence inspection robot use in a large number, contain abundant power equipment health status in the video image, nevertheless because intelligent technical level limits, data such as a large amount of video image that produce in the operation process do not carry out the degree of depth and excavate, do not realize yet with the degree of depth integration of distribution thing networking monitoring system and merge, the urgent need artificial intelligence technique supports.
Disclosure of Invention
The embodiment of the invention aims to provide a monitoring method, a monitoring device and a monitoring system for a power distribution network.
In order to achieve the above object, an embodiment of the present invention provides a method for monitoring a power distribution network, where the method for monitoring the power distribution network includes: acquiring state information in the power distribution network in real time, wherein the state information is at least one of a state video vector, a state image vector and a state text vector; taking the state information as particles to be brought into a PSO particle swarm algorithm to determine optimal parameters of the particles; bringing the optimal parameters of the particles into a BilSTM model to obtain a key feature set; determining attention degrees of different features in the key feature set by combining an AM algorithm and the BilSTM model; monitoring for faults in the power distribution network based on the attentiveness of the characteristic.
Optionally, the PSO particle swarm algorithm is to dynamically adjust a learning factor:
wherein the content of the first and second substances,for the distance between each particle and the current global optimum,andin order to improve the learning factor of the learning factor,as a learning factorThe maximum value of (a) is,as a learning factorThe minimum value of (a) is determined,is the maximum value of the distance of the particle from the global optimum,in order to be able to perform the number of iterations,in order to be the maximum number of iterations,is the position of the particle after t iterations, j is the dimension of the search space,is the j-dimensional value of the global optimal position.
Optionally, the bringing the optimal parameters of the particles into the BiLSTM model to obtain a key feature set includes: taking the optimal parameters of the particles as input vectors to be brought into a BilSTM model to obtain a key feature set, wherein the BilSTM model comprises the following steps:
wherein the content of the first and second substances,in order to forget to leave the door,in order to memorize the door, the door is provided with a memory,in order to be the current memory state,in order to remember the memory state at the last moment,in order to temporarily memorize the state of the memory,in order to activate the function(s),in order to be a weight matrix, the weight matrix,for the input vector at this moment, m is the dimension of the input vectorN is the number of the hidden layer units,is a hidden state at the last moment,in order to output the output gate, the output gate is provided with a gate,in order to be a vector of the offset,and the current hidden state is a set of the hidden states as a key feature set.
Optionally, the BiLSTM model simplifies the current memory state through the memory gate.
Optionally, the determining the attention degree of different features in the key feature set by combining the AM algorithm and the BiLSTM model includes: determining the degree of attention of different features in the set of key features by the formula:
wherein the content of the first and second substances,the attention weight is expressed for the feature,is a hidden state output by the BilSTM model,in order to be a probability vector, the probability vector,the weighting value is a degree of attention to different features in the set of key features.
Optionally, the monitoring faults in the power distribution network according to the attention degree of the features includes: when the attention degree of the characteristics is smaller than the fault threshold range, the power distribution network stops monitoring the state information corresponding to the characteristics; and when the attention degree of the characteristics is not less than the fault threshold range, the power distribution network starts monitoring the state information corresponding to the characteristics and starts fault alarm.
In another aspect, the present invention provides a monitoring device for a power distribution network, the monitoring device comprising: the acquiring unit is used for acquiring state information in the power distribution network, wherein the state information is at least one of a state video vector, a state image vector and a state text vector; the optimization unit takes the state information as particles and brings the state information into a PSO particle swarm algorithm to determine optimal parameters of the particles; the first processing unit is used for bringing the optimal parameters of the particles into a BilSTM model to obtain a key feature set; a second processing unit, configured to combine the AM algorithm and the BiLSTM model to determine attention degrees of different features in the key feature set; and the management unit is used for monitoring faults in the power distribution network according to the attention degree.
Optionally, the PSO particle swarm algorithm is to dynamically adjust a learning factor:
wherein the content of the first and second substances,for the distance between each particle and the current global optimum,
andin order to improve the learning factor,as a learning factorThe maximum value of (a) is,as a learning factorThe minimum value of (a) is determined,is the maximum value of the distance of the particle from the global optimum,in order to be able to perform the number of iterations,in order to be the maximum number of iterations,is the position of the particle after t iterations, j is the dimension of the search space,is the j-dimensional value of the global optimal position.
Optionally, the bringing the optimal parameters of the particles into the BiLSTM model to obtain a key feature set includes: taking the optimal parameters of the particles as input vectors to be brought into a BilSTM model to obtain a key feature set, wherein the BilSTM model comprises the following steps:
wherein the content of the first and second substances,in order to forget to leave the door,in order to memorize the door, the door is provided with a memory,in order to obtain the current memory state,in order to remember the memory state at the last moment,in order to temporarily memorize the state of the memory,in order to activate the function(s),in order to be a weight matrix, the weight matrix,is the input vector at this moment, m is the dimension of the input vector, n is the number of hidden layer units,is a hidden state at the last moment,in order to output the output gate, the output gate is provided with a gate,in order to be a vector of the offset,and the current hidden state is a set of the hidden states as a key feature set.
Optionally, the BiLSTM model simplifies the current memory state through the memory gate.
Optionally, the determining the attention degree of different features in the key feature set by combining the AM algorithm and the BiLSTM model includes: determining the degree of attention of different features in the set of key features by the following formula:
wherein the content of the first and second substances,the attention weight is expressed for a feature,for the hidden state output by the BiLSTM model,in order to be a probability vector, the probability vector,the weighting value is a degree of attention to different features in the set of key features.
Optionally, when the attention degree of the feature is smaller than the fault threshold range, the power distribution network closes monitoring on the state information corresponding to the feature; and when the attention degree of the characteristics is not less than the fault threshold range, the power distribution network starts monitoring the state information corresponding to the characteristics and starts fault alarm.
On the other hand, the invention also provides a monitoring system of the power distribution network, which comprises the monitoring device of the power distribution network and at least one terminal, wherein the monitoring device is used for monitoring the terminal.
The monitoring method of the power distribution network comprises the following steps: acquiring state information in the power distribution network in real time, wherein the state information is at least one of a state video vector, a state image vector and a state text vector; taking the state information as particles into a PSO particle swarm algorithm for determining optimal parameters of the particles; substituting the particle optimal parameters into a BilSTM model to obtain a key feature set; combining an AM algorithm and the BilSTM model for determining attention degrees of different features in the key feature set; monitoring for faults in the power distribution network based on the attentiveness of the characteristic. The monitoring method improves the accuracy and convergence time of power distribution network monitoring fault positioning, realizes the flow control of the power distribution terminal equipment fault, and reduces the operation risk of the terminal equipment.
Additional features and advantages of embodiments of the invention will be set forth in the detailed description which follows.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the embodiments of the invention without limiting the embodiments of the invention. In the drawings:
fig. 1 is a schematic flow chart of a monitoring method for a power distribution network according to the present invention;
FIG. 2 is a schematic flow diagram of the IPSO-IBiLSTM-AM architecture of the present invention;
FIG. 3 is a hierarchical schematic of the IBiLSTM-AM algorithm of the present invention.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating embodiments of the invention, are given by way of illustration and explanation only, not limitation.
Fig. 1 is a schematic flow chart of a monitoring method for a power distribution network according to the present invention, and as shown in fig. 1, step S101 is to obtain state information in the power distribution network in real time, where the state information is at least one of a state video vector, a state image vector, and a state text vector. Specifically, the state information is obtained by performing vectorization conversion on a state video image and/or a text to obtain a vector expression of the vector information, where the vector expression includes a speed attribute and a position attribute of the state information.
And step S102, taking the state information as particles into a PSO particle swarm algorithm for determining optimal parameters of the particles. The PSO Particle Swarm Optimization (PSO) is a random search Algorithm based on group cooperation, the principle of which is derived from the research on the predation behavior of a bird group, and the basic idea is to find the optimal solution through cooperation and information sharing among individuals in the group. The PSO algorithm is generated by the influence of the behavior of the bird swarm to search for food, and there are 2 factors that influence the bird search for food in the bird swarm: the speed at which the bird flies, the distance between the bird and the food. Likewise, the birds in the group of birds are considered as particles in the group of particles, and the process of searching for food is considered as the process of searching for the optimal solution by the particles. The particle optimal parameter is the optimal solution of the particle search.
Each particle has 2 attributes of speed and position, in each iterative calculation, each particle searches for an optimal solution at a certain speed and position in a K-dimensional space, and the position of the searched optimal solution is marked as an individual extreme value.
The PSO particle swarm algorithm comprises the following basic processes: randomly initializing each particle; evaluating each particle and obtaining global optimum; judging whether the global optimum meets an end condition, if so, ending, and if not, updating the speed and the position of each particle; evaluating a function fitness value for each particle; updating the historical optimal position of each particle; and updating the global optimal position of the population.
Specifically, the standard particle swarm algorithm has the following flow:
1) setting parameters such as population size, variable range, inertial weight, learning factors and the like, and randomly initializing a group of particles (including speed and position information) uniformly distributed in a given optimization space;
2) calculating the fitness value (namely a function value) of each particle in the population, setting the fitness value of the ith particle as the current individual extreme value of the ith particle, and setting the best particle in all the particles as the overall extreme value of the population;
3) updating the velocity and position of each particle according to equations 1) and 2);
4) comparing the current function value of all the particles with the best position found previously, if the current position is better, setting the individual optimal position as the position of the particle, and then updating the global extreme value of the group;
5) judging whether a given termination 1 condition is met, if so, stopping searching and outputting a required result; otherwise, return to 3) to continue searching.
The correlation formula is as follows: assuming an objective function of,For the ith particle position, the ith particle individual extremum is expressed as:
assuming that the particle swarm contains N particles, the iteration number is t, and the global optimal position isThe solving process is as follows:
Each particle adjusts its position and velocity by a global optimum position. Velocity of the particlesAnd positionThe calculation formula is as follows:
wherein: i is the number of particles in the particle group, i is 1, 2, … N; j is the dimension of the search space, j is 1, 2, … K;the velocity of the particles after t iterations;the position of the particle after t iterations;j-dimensional value of the individual extreme value of the ith particle;a j-dimensional value of the global optimal position;andis [0, 1 ]]A random number above;andis a learning factor;for inertial weight, by adjustmentThe global optimization performance and the local optimization performance can be adjusted, and particularly,andalso referred to as the acceleration constant, of the,for each particle an individual learning factor is used,social learning factor for each particle, in generalAndis a constant of 0 to 4.
Obtaining the fitness value of the particle according to the following formula, and determining the global optimal position of the particle according to the fitness value:
wherein:is a value of the fitness of the particle,is the true value of the particle or particles,is the predicted value of the particle, and n is the sample number of the particle.
The invention introduces the distance between each particle and the current global optimal position in the t iterationBy usingThe invention has the characteristics of larger learning factor in the early period of optimization and smaller learning factor in the later period of optimizationAndthe improvement is carried out, and the learning factor is dynamically adjusted, wherein:
wherein the content of the first and second substances,as a learning factorThe maximum value of (a) is,as a learning factorThe minimum value of (a) is determined,is the maximum value of the distance of the particle from the global optimum,is the maximum number of iterations. By pairsAnda non-linear dynamic adjustment is carried out so thatIndividual whose value determines the particleThe ability of the user to recognize the position of the user,the social cognitive ability of the particles is determined, and the convergence speed and precision of the PSO algorithm are improved.
The hyper-parameter Optimization part of the invention automatically finds the hyper-parameter combination which enables the fault location effect to be optimal by utilizing an Improved PSO Algorithm (also called IPSO Algorithm: Improved Particle Swarm Optimization Algorithm), and the IPSO Algorithm introduces the distance between each Particle and the current global optimal position on the basis of a standard PSO Algorithm to optimize the learning factor and realize dynamic adjustment.
Step S103, substituting the optimal parameters of the particles into a BilSTM (Bi-directional Long Short-Term Memory model) to obtain a key feature set; the BilSTM model is formed by combining a forward LSTM (Long Short-Term Memory model) and a backward LSTM. Both are often used to model context information in natural language processing tasks, LSTM is an improved model of RNN (current Neural Network Recurrent Neural Network), where the standard BiLSTM algorithm Neural Network is: let m be the dimension of the input vector, n be the number of hidden layer units, according to the hidden state of the last momentAnd input at this momentGet the forget doorMemory doorAnd output gateFromAnd memory state at last momentObtaining the current memory stateFromAndobtaining the hidden state at the current moment. Forgetting, memorizing and outputting of the memory unit are achieved, important information is reserved, information with low importance degree is ignored, key feature identification of input service flow can be achieved, and a key feature set is output. The specific formula is as follows:
wherein, the first and the second end of the pipe are connected with each other,in order to activate the function(s),
、、in order to be a weight matrix, the weight matrix,、in order to be a vector of the offset,in order to temporarily memorize the state of the memory,is the input vector at this moment.
The BilSTM is composed of forward and reverse LSTMs, and the hidden state isSaid hidden stateIs a key feature set:
wherein, the first and the second end of the pipe are connected with each other,is in the forward direction,Is in the reverse direction。
The invention provides an Improved BilSTM algorithm neural network, namely IBiLSTM (Improved Bi-directional Long Short-Term Memory), which specifically comprises the steps of simplifying the gate structure and parameters of standard LSTM, and using the gate structure and parametersSimplifying the current memory state and removing the input weightAnd a bias matrixThe LSTM algorithm is improved, and the training time of the network is shortened on the premise of not losing the precision. And taking the optimal parameters of the particles as input vectors to be introduced into an improved BilSTM model, wherein the improved BilSTM model has the following specific formula:
wherein, the first and the second end of the pipe are connected with each other,in order to forget to leave the door,in order to memorize the door, the door is provided with a memory,in order to be the current memory state,in order to remember the memory state at the last moment,in order to temporarily memorize the state of the memory,in order to activate the function(s),in the form of a matrix of weights,is the input vector at this moment, m is the dimension of the input vector, n is the number of hidden layer units,is a hidden state at the last moment,in order to output the output gate, the output gate is provided with a gate,in order to be a vector of the offset,and the current hidden state is a set of the hidden states as a key feature set.
The difference between the improved BilSTM algorithm designed by the invention and the control signal of the standard neural network is to forget the gateInstead, by memory gatesAnd an output gateTwo gate structures are formed so that the network gate structure signal is output by a t-1 moment output signalThe recursive weight matrix and the bias matrix, and the calculation complexity is reduced.
Step S104 is to combine the AM algorithm and the BiLSTM model for determining attention degrees of different features in the key feature set.
The AM algorithm (attention model) is an attention mechanism algorithm, a BilSTM-AM (Bi-directional Long Short-Term Memory and attention model) identifies an attention mechanism, the attention mechanism is introduced into the BilSTM model, calculation with probability distribution is obtained through calculation of attention probability distribution, attention degrees of different characteristics in fault description are highlighted, and the purpose of improving classification accuracy is achieved.
Hidden states output from a BilSTM modelFormula for feature expression attention weightComprises the following steps:
probability calculation is carried out by utilizing a softmax function to obtain a probability vector, and the hidden state is multiplied by the corresponding probability vector to obtain a weighted value formula:
wherein the content of the first and second substances,the attention weight is expressed for the feature,for the hidden state output by the BiLSTM model,in order to be a probability vector, the probability vector,the weighting value is a degree of attention to different features in the set of key features.
Step S105 is monitoring a fault in the power distribution network according to the attention degree of the feature, including: when the attention degree of the characteristic is smaller than the fault threshold range, the power distribution network stops monitoring the state information corresponding to the characteristic; and when the attention degree of the characteristics is not less than the fault threshold range, the power distribution network starts monitoring the state information corresponding to the characteristics and starts fault alarm. The fault threshold range is determined according to methods such as experience and big data test.
FIG. 2 is a schematic flow diagram of the IPSO-IBiLSTM-AM construction of the present invention, and as shown in FIG. 2, the present invention fuses the proposed hyper-parameter optimization part and the fault localization part to construct an IPSO-IBiLSTM-AM algorithm, the construction process is as follows: configuring a fault location part by using the initial hyper-parameter provided by the hyper-parameter optimization part; the hyper-parameter optimization part optimizes the hyper-parameters according to the output result of the iterative operation of the prediction part.
The specific training steps of the IPSO-IBiLSTM-AM are as follows:
step 1: dividing a data set into a training set and a test set according to a proportion, and carrying out vectorization processing;
step 2: initializing parameters such as the number of units, the speed, the position, the inertia weight, the acceleration factor, the iteration times and the like of a BilSTM layer;
and step 3: and (3) training the model part by using a training set, and solving the individual optimal position and the global position according to the fitness value of the particles, wherein the fitness calculation formula is as follows:
wherein:in order to be the true value of the value,is a predicted value, n is the number of samples;
and 4, step 4: according toAnddynamically adjusting the learning factor by a formula, and starting iteration;
and 5: and (4) judging whether the termination condition is met, if so, outputting the optimal position, and otherwise, returning to the step (3). The termination condition is that the particle fitness is smaller than a set value or the maximum iteration number is reached;
step 6: after the operation is finished, distributing the result obtained by IPSO to an IBiLSTM-AM neural network as an optimal hyper-parameter combination, and training the combination by using a training set;
and 7: inputting the characteristic data of the test set into the trained model, and outputting a fault positioning result.
FIG. 3 is a schematic diagram of an IBiLSTM-AM (attribute model and Improved Bi-directional Short-Term Memory) algorithm layer, as shown in FIG. 3, the hyper-parameter optimization part combines an attention mechanism with a neural network model, and comprises a series of different layers, namely an input layer, an IBiLSTM layer, an AM layer and a full connection layer. The layers function as follows:
an input layer: the feature data at each moment is used as the input of the model, and the state (including fault state) video image and text are required to be subjected to vectorization conversion to obtain the vector expression of the state information.
IBiLSTM layer: and (3) performing feature extraction on the fault description word vector by using two simplified LSTM neural networks in opposite directions to obtain semantic features of fault description.
An AM layer: and the attention mechanism layer is used for carrying out weight distribution on the semantic features, improving the weight of important information in the semantic features and obtaining a final description feature vector of the fault.
Full connection layer: the full-connection layer is used for reducing dimensionality operation and facilitating classification and identification, and specifically, a sigmoid (S-shaped growth curve) activation function is selected as a gating mechanism to classify the output of the attention layer to obtain a fault type result.
The invention also provides a monitoring device of the power distribution network, which comprises: the acquiring unit is used for acquiring state information in the power distribution network, wherein the state information is at least one of a state video vector, a state image vector and a state text vector; the optimization unit takes the state information as particles into a PSO particle swarm algorithm and is used for determining optimal parameters of the particles; the first processing unit is used for bringing the optimal parameters of the particles into a BilSTM model to obtain a key feature set; a second processing unit for combining an AM algorithm with the BilSTM model to determine attention degrees of different features in the key feature set; and the management unit is used for monitoring faults in the power distribution network according to the attention degree.
On the other hand, the invention also provides a monitoring system of the power distribution network, which comprises the monitoring device of the power distribution network and at least one terminal, wherein the monitoring device is used for monitoring the terminal.
The monitoring method of the power distribution network comprises the following steps: acquiring state information in the power distribution network in real time, wherein the state information is at least one of a state video vector, a state image vector and a state text vector; taking the state information as particles into a PSO particle swarm algorithm for determining optimal parameters of the particles; bringing the optimal parameters of the particles into a BilSTM model to obtain a key feature set; combining an AM algorithm and the BilSTM model to determine attention degrees of different features in the key feature set; monitoring for faults in the power distribution network based on the attentiveness of the characteristic. The monitoring method not only improves the door structure and parameters and simplifies the standard BilSTM model, but also dynamically adjusts the learning factors in the PSO particle swarm algorithm, improves the accuracy and convergence time of the monitoring fault positioning of the power distribution network, realizes the flow management and control of the faults of the power distribution terminal equipment, and reduces the operation risk of the terminal equipment.
Although the embodiments of the present invention have been described in detail with reference to the accompanying drawings, the embodiments of the present invention are not limited to the details of the above embodiments, and various simple modifications can be made to the technical solutions of the embodiments of the present invention within the technical idea of the embodiments of the present invention, and the simple modifications all belong to the protection scope of the embodiments of the present invention.
It should be noted that the various features described in the foregoing embodiments may be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention do not describe every possible combination.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in the process, method, article, or apparatus that comprises the element.
The above are merely examples of the present application and are not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement or the like made within the spirit and principle of the present application shall be included in the scope of the claims of the present application.
Claims (9)
1. A monitoring method for a power distribution network is characterized by comprising the following steps:
acquiring state information in the power distribution network in real time, wherein the state information is at least one of a state video vector, a state image vector and a state text vector;
taking the state information as particles to be brought into a PSO particle swarm algorithm to determine optimal parameters of the particles;
bringing the optimal parameters of the particles into a BilSTM model to obtain a key feature set;
determining attention degrees of different features in the key feature set by combining an AM algorithm and the BilSTM model;
monitoring faults in the power distribution network according to the attention degree of the characteristics;
the PSO particle swarm algorithm is a dynamic regulation learning factor:
wherein the content of the first and second substances,for the distance between each particle and the current global optimum,
andin order to improve the learning factor of the learning factor,as a learning factorThe maximum value of (a) is,as a learning factorThe minimum value of (a) is determined,
is the maximum value of the distance of the particle from the global optimum,in order to be the number of iterations,in order to be the maximum number of iterations,
a j-dimensional value of the global optimal position; the monitoring for faults in the power distribution network according to the degree of attention of the feature comprises:
when the attention degree of the characteristics is smaller than the fault threshold range, the power distribution network stops monitoring the state information corresponding to the characteristics;
and when the attention degree of the characteristics is not less than the fault threshold range, the power distribution network starts monitoring the state information corresponding to the characteristics and starts fault alarm.
2. The monitoring method of claim 1, wherein the bringing the particle optimal parameters into a BilSTM model to obtain a set of key features comprises:
taking the optimal parameters of the particles as input vectors to be brought into a BilSTM model to obtain a key feature set, wherein the BilSTM model comprises the following steps:
wherein the content of the first and second substances,in order to forget to leave the door,in order to memorize the door, the door is provided with a memory,
in order to obtain the current memory state,in order to remember the memory state at the last moment,in order to temporarily memorize the state of the memory,
m is the dimension of the input vector, n is the number of hidden layer units,
is a hidden state at the last moment,in order to output the gate, the gate is provided with a gate,in order to be a vector of the offset,
3. The monitoring method according to claim 2,
and the BilSTM model simplifies the current memory state through the memory gate.
4. The method of monitoring of claim 1, wherein said determining a degree of attention for different features in said set of key features in combination with an AM algorithm and said BiLSTM model comprises:
determining the degree of attention of different features in the set of key features by the following formula:
wherein the content of the first and second substances,the attention weight is expressed for the feature,
for the hidden state output by the BiLSTM model,in order to be a probability vector, the probability vector,
the weighting value, which is the degree of attention of the different features in the key feature set,
softmax is a probabilistic calculation function.
5. A monitoring device for an electrical distribution network, the monitoring device comprising:
the acquiring unit is used for acquiring state information in the power distribution network, wherein the state information is at least one of a state video vector, a state image vector and a state text vector;
the optimization unit takes the state information as particles to be brought into a PSO particle swarm algorithm so as to determine optimal parameters of the particles;
the first processing unit is used for bringing the optimal parameters of the particles into a BilSTM model to obtain a key feature set;
a second processing unit, configured to combine the AM algorithm and the BiLSTM model to determine attention degrees of different features in the key feature set;
the management unit is used for monitoring faults in the power distribution network according to the attention degree;
the PSO particle swarm algorithm is a dynamic regulation learning factor:
wherein the content of the first and second substances,for the distance between each particle and the current global optimum,
andin order to improve the learning factor of the learning factor,as a learning factorThe maximum value of (a) is,as a learning factorThe minimum value of (a) is calculated,
in order to be able to perform the number of iterations,is the maximum number of iterations in the sequence,
the monitoring for faults in the power distribution network according to the degree of attention of the feature comprises:
when the attention degree of the characteristic is smaller than the fault threshold range, the power distribution network stops monitoring the state information corresponding to the characteristic;
and when the attention degree of the characteristics is not less than the fault threshold range, the power distribution network starts monitoring the state information corresponding to the characteristics and starts fault alarm.
6. The monitoring device of claim 5, wherein the substituting the particle optimal parameters into the BilSTM model to obtain a set of key features comprises:
taking the optimal parameters of the particles as input vectors to be brought into a BilSTM model to obtain a key feature set, wherein the BilSTM model comprises the following steps:
in order to be the current memory state,in order to remember the memory state at the last moment,in order to temporarily memorize the state of the memory,
m is the dimension of the input vector, n is the number of hidden layer units,
is a hidden state at the last moment,in order to output the gate, the gate is provided with a gate,in order to be a vector of the offset,
7. The monitoring device of claim 6,
and the BilSTM model simplifies the current memory state through the memory gate.
8. The monitoring device of claim 5, wherein said determining the degree of attention of different features in said set of key features in combination with an AM algorithm and said BilStm model comprises:
determining the degree of attention of different features in the set of key features by the following formula:
wherein the content of the first and second substances,the attention weight is expressed for the feature,
the weighted value is the attention degree of different characteristics in the key characteristic set;
softmax is a probabilistic calculation function.
9. A monitoring system for an electric distribution network, characterized in that the monitoring system comprises a monitoring device for an electric distribution network according to any of claims 5-8 and at least one terminal, said monitoring device being adapted to monitor said terminal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210846447.5A CN114915035B (en) | 2022-07-19 | 2022-07-19 | Power distribution network monitoring method, device and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210846447.5A CN114915035B (en) | 2022-07-19 | 2022-07-19 | Power distribution network monitoring method, device and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114915035A CN114915035A (en) | 2022-08-16 |
CN114915035B true CN114915035B (en) | 2022-09-13 |
Family
ID=82772913
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210846447.5A Active CN114915035B (en) | 2022-07-19 | 2022-07-19 | Power distribution network monitoring method, device and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114915035B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111429034A (en) * | 2020-04-21 | 2020-07-17 | 国网信通亿力科技有限责任公司 | Method for predicting power distribution network fault |
AU2020101683A4 (en) * | 2020-08-05 | 2020-09-10 | Abu-Siada, Ahmed ASSOC PROF | Fault detection, location, and prediction within an electricity power transmission and distribution networks |
CN113327033A (en) * | 2021-05-28 | 2021-08-31 | 广西电网有限责任公司电力科学研究院 | Power distribution network fault diagnosis method and system |
CN113484669A (en) * | 2021-06-23 | 2021-10-08 | 国网江苏省电力有限公司淮安供电分公司 | Bidirectional LSTM-based power distribution network low-voltage reason positioning method |
CN113964885A (en) * | 2021-08-31 | 2022-01-21 | 国网山东省电力公司东营供电公司 | Reactive active prediction and control technology of power grid based on situation awareness |
-
2022
- 2022-07-19 CN CN202210846447.5A patent/CN114915035B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111429034A (en) * | 2020-04-21 | 2020-07-17 | 国网信通亿力科技有限责任公司 | Method for predicting power distribution network fault |
AU2020101683A4 (en) * | 2020-08-05 | 2020-09-10 | Abu-Siada, Ahmed ASSOC PROF | Fault detection, location, and prediction within an electricity power transmission and distribution networks |
CN113327033A (en) * | 2021-05-28 | 2021-08-31 | 广西电网有限责任公司电力科学研究院 | Power distribution network fault diagnosis method and system |
CN113484669A (en) * | 2021-06-23 | 2021-10-08 | 国网江苏省电力有限公司淮安供电分公司 | Bidirectional LSTM-based power distribution network low-voltage reason positioning method |
CN113964885A (en) * | 2021-08-31 | 2022-01-21 | 国网山东省电力公司东营供电公司 | Reactive active prediction and control technology of power grid based on situation awareness |
Also Published As
Publication number | Publication date |
---|---|
CN114915035A (en) | 2022-08-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Qolomany et al. | Parameters optimization of deep learning models using particle swarm optimization | |
Khajavi et al. | Predicting the carbon dioxide emission caused by road transport using a Random Forest (RF) model combined by Meta-Heuristic Algorithms | |
CN111260124A (en) | Chaos time sequence prediction method based on attention mechanism deep learning | |
CN113489654B (en) | Routing method, device, electronic equipment and storage medium | |
Nagahamulla et al. | An ensemble of artificial neural networks in rainfall forecasting | |
CN111416797A (en) | Intrusion detection method for optimizing regularization extreme learning machine by improving longicorn herd algorithm | |
CN115511177A (en) | Ultra-short-term wind speed prediction method based on INGO-SWGMN hybrid model | |
CN113722980A (en) | Ocean wave height prediction method, system, computer equipment, storage medium and terminal | |
Froelich et al. | Predictive capabilities of adaptive and evolutionary fuzzy cognitive maps-a comparative study | |
Culclasure | Using neural networks to provide local weather forecasts | |
CN115168720A (en) | Content interaction prediction method and related equipment | |
CN112770256B (en) | Node track prediction method in unmanned aerial vehicle self-organizing network | |
CN113825165B (en) | 5G slice network congestion early warning method and device based on time diagram network | |
CN114915035B (en) | Power distribution network monitoring method, device and system | |
Chen et al. | A Spark-based Ant Lion algorithm for parameters optimization of random forest in credit classification | |
Shi et al. | A modified particle swarm optimization and radial basis function neural network hybrid algorithm model and its application | |
Li et al. | Prediction of short-term traffic flow based on PSO-optimized chaotic BP neural network | |
Lima et al. | Cognitive smart cities and deep learning: A classification framework | |
Iizaka et al. | A novel daily peak load forecasting method using analyzable structured neural network | |
CN114548297A (en) | Data classification method, device, equipment and medium based on domain self-adaption | |
CN114372418A (en) | Wind power space-time situation description model establishing method | |
Bin et al. | Simple and effective fault diagnosis method of power lithium-ion battery based on GWA-DBN | |
Liu et al. | Wind power prediction based on LSTM-CNN optimization | |
Li et al. | A novel RBF neural network design based on immune algorithm system | |
Anh et al. | Wind speed intervals prediction using meta-cognitive approach |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |