CN117709712A - Situation prediction method and terminal for power distribution network based on hybrid neural network - Google Patents

Situation prediction method and terminal for power distribution network based on hybrid neural network Download PDF

Info

Publication number
CN117709712A
CN117709712A CN202311652456.1A CN202311652456A CN117709712A CN 117709712 A CN117709712 A CN 117709712A CN 202311652456 A CN202311652456 A CN 202311652456A CN 117709712 A CN117709712 A CN 117709712A
Authority
CN
China
Prior art keywords
data
power distribution
distribution network
neural network
hybrid neural
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
Application number
CN202311652456.1A
Other languages
Chinese (zh)
Inventor
方朝雄
张林垚
郑洁云
周霞
陈卓琳
张章煌
刘贵宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Fujian Electric Power Co Ltd
Economic and Technological Research Institute of State Grid Fujian Electric Power Co Ltd
Original Assignee
State Grid Fujian Electric Power Co Ltd
Economic and Technological Research Institute of State Grid Fujian Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by State Grid Fujian Electric Power Co Ltd, Economic and Technological Research Institute of State Grid Fujian Electric Power Co Ltd filed Critical State Grid Fujian Electric Power Co Ltd
Priority to CN202311652456.1A priority Critical patent/CN117709712A/en
Publication of CN117709712A publication Critical patent/CN117709712A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Economics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Educational Administration (AREA)
  • Primary Health Care (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a situation prediction method and a terminal of a power distribution network based on a hybrid neural network, which are characterized in that field operation data of the power distribution network are acquired through a virtual acquisition technology, missing data in the acquired data set are subjected to interpolation processing through an interpolation method, noise in original data is removed, and then feature selection dimension reduction processing is performed on the processed data; and optimizing three super parameters, namely hidden layer dimension, maximum training period and initial learning rate, of the hybrid neural network, and finally inputting processed data into the optimized hybrid neural network to predict the situation of the power distribution network. In this way, the operation risk of the intelligent power grid can be estimated in real time, the future change trend of the state of the intelligent power grid can be accurately predicted, and the distribution management department can be timely reminded to make safety regulation.

Description

Situation prediction method and terminal for power distribution network based on hybrid neural network
Technical Field
The invention relates to the technical field of situation prediction of power distribution networks, in particular to a situation prediction method and a terminal of a power distribution network based on a hybrid neural network.
Background
With the development of the intelligent power grid, the power distribution network is mixed with alternating current and direct current, and the source-network-charge-storage cross coupling and the multisource complementation are carried out, so that the running state of the intelligent power distribution network is increasingly complex, and the instability risk is increasingly increased; in addition, the high-proportion grid connection of clean energy and the disordered access of distributed generation cause the remarkable increase of the uncertainty of the power grid, and the accurate prediction of the future change trend of the state of the intelligent power grid becomes one of the important concerns in the field of the power system in the current era.
The situation prediction is a state prediction stage and is used for predicting the future change trend of the state of the smart grid. Meanwhile, the situation prediction can monitor the running state of the power distribution network on line in real time, discover the running risk of the power distribution network in time and remind a distribution network management department of making adjustments in time. The situation prediction technology mainly comprises the following steps: the uncertainty distributed power output technology, the intelligent power distribution network safety analysis and early warning technology, the three-phase unbalanced load prediction technology, the load layering and grading technology and the electric vehicle charging load prediction technology based on big data are considered.
The intelligent power distribution network is used as a hub of the power energy system, is directly oriented to the user terminal, and the stability of the intelligent power distribution network directly relates to whether a terminal user can safely and reliably use electricity, and can not predict and analyze the situation of the intelligent power distribution network. However, the current smart grid situation prediction technology faces the difficulty of accurately predicting the future change trend of the power grid state by evaluating the running risk of the power grid in real time.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the situation prediction method and the terminal for the power distribution network based on the hybrid neural network can evaluate the running risk of the power grid in real time and accurately predict the future change trend of the power grid state.
In order to solve the technical problems, the invention adopts the following technical scheme:
a situation prediction method of a power distribution network based on a hybrid neural network comprises the following steps:
the method comprises the steps of collecting field operation data of a power distribution network by using virtual collection;
interpolation processing is carried out on the missing data in the acquired data set, denoising processing is carried out on the acquired data set, and feature selection and dimension reduction processing are carried out on the interpolated data and the denoised data;
optimizing the hidden layer dimension, the maximum training period and the initial learning rate of the hybrid neural network;
and inputting the dimensionality-reduced processed data into the optimized mixed neural network, and predicting the situation of the power distribution network.
In order to solve the technical problems, the invention adopts another technical scheme that:
the situation prediction terminal of the power distribution network based on the hybrid neural network comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the processor realizes the steps of the situation prediction method of the power distribution network based on the hybrid neural network when executing the computer program.
The invention has the beneficial effects that: acquiring field operation data of the power distribution network through a virtual acquisition technology, performing interpolation processing on missing data in the acquired data set through an interpolation method, removing noise in original data, and performing feature selection dimension reduction processing on the processed data; and optimizing three super parameters, namely hidden layer dimension, maximum training period and initial learning rate, of the hybrid neural network, and finally inputting processed data into the optimized hybrid neural network to predict the situation of the power distribution network. In this way, the operation risk of the intelligent power grid can be estimated in real time, the future change trend of the state of the intelligent power grid can be accurately predicted, and the distribution management department can be timely reminded to make safety regulation.
Drawings
Fig. 1 is a flowchart of a situation prediction method of a power distribution network based on a hybrid neural network according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a situation prediction terminal of a power distribution network based on a hybrid neural network according to an embodiment of the present invention;
FIG. 3 is a flow chart of feature selection in accordance with an embodiment of the present invention;
fig. 4 is a block diagram of a CNN-BILSTM hybrid neural network according to an embodiment of the present invention.
Description of the reference numerals:
1. situation prediction terminal of power distribution network based on hybrid neural network; 2. a memory; 3. a processor.
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.
Referring to fig. 1, an embodiment of the present invention provides a situation prediction method for a power distribution network based on a hybrid neural network, including the steps of:
the method comprises the steps of collecting field operation data of a power distribution network by using virtual collection;
interpolation processing is carried out on the missing data in the acquired data set, denoising processing is carried out on the acquired data set, and feature selection and dimension reduction processing are carried out on the interpolated data and the denoised data;
optimizing the hidden layer dimension, the maximum training period and the initial learning rate of the hybrid neural network;
and inputting the dimensionality-reduced processed data into the optimized mixed neural network, and predicting the situation of the power distribution network.
From the above description, the beneficial effects of the invention are as follows: acquiring field operation data of the power distribution network through a virtual acquisition technology, performing interpolation processing on missing data in the acquired data set through an interpolation method, removing noise in original data, and performing feature selection dimension reduction processing on the processed data; and optimizing three super parameters, namely hidden layer dimension, maximum training period and initial learning rate, of the hybrid neural network, and finally inputting processed data into the optimized hybrid neural network to predict the situation of the power distribution network. In this way, the operation risk of the intelligent power grid can be estimated in real time, the future change trend of the state of the intelligent power grid can be accurately predicted, and the distribution management department can be timely reminded to make safety regulation.
Further, the interpolating processing for the missing data in the collected data set, and the denoising processing for the collected data set, includes:
and carrying out interpolation processing on the missing data in the data set of the collected field operation data by using cubic spline interpolation, and removing noise in the data set of the collected field operation data by adopting continuous wavelet transformation.
From the above description, it can be seen that the missing data in the collected data set is interpolated by using a cubic spline interpolation method, so that information loss can be avoided, and noise in the collected data set is removed by using continuous wavelet transformation, so that the quality of the data set can be further improved.
Further, the interpolation processing is performed on the missing data in the collected data set of the field operation data by using a cubic spline interpolation method, including:
calculating interpolation step h i ＝x i+1 -x i (i=1, 2, …, n), n representing the number of data in the data set, and the data node being represented as (x) 1 ,y 1 ),(x 2 ,y 2 ),(x 3 ,y 3 ),…,(x n ,y n )；
Calculating coefficient a of spline curve i ,b i ,c i ,d i ：
In each subinterval x i ≤x≤x i+1 In g i (x)＝a i +b i (x-x i )+c i (x-x i ) 2 +d i (x-x i ) 3 ；
In the formula g i (x) Interpolation data representing the missing value at i.
Further, the feature selection of the interpolated data and the denoised data includes:
and adopting L1 norm to perform feature selection on the interpolated data and the denoised data.
From the above description, since the L1 norm refers to the sum of absolute values of the elements in the vector, in feature selection, the L1 norm is used as a regularization term, which can cause some feature weights in the model to become 0, thereby realizing feature sparsification. Therefore, by sparsifying the feature weights, the L1 norm feature selection can exclude the features which do not contribute to the prediction task, and the generalization capability and the interpretation of the model are improved.
Further, the dimension reduction processing is carried out on the data, which comprises the following steps:
computing a kernel matrix between the data:
wherein K (x i ,x j ) Representing datax i And x j The result of the kernel function calculation between them,respectively represent data x i And x j Feature vectors mapped to a high-dimensional space;
centering the core matrix:
and decomposing the characteristic values of the centralized kernel matrix to obtain characteristic vectors and characteristic values, sorting the characteristic values in a descending order according to the sizes of the characteristic values, selecting the front preset characteristic values and the corresponding characteristic vectors after sorting as main components, and projecting each sample in a high-dimensional data set through the main components to obtain a corresponding low-dimensional representation.
As can be seen from the above description, by the nonlinear dimension reduction method based on the kernel method, the high-dimensional data can be mapped to the low-dimensional space, so as to extract important information in the data and improve the accuracy of the subsequent situation prediction.
Further, optimizing the hidden layer dimension, the maximum training period and the initial learning rate of the hybrid neural network comprises optimizing by adopting a dung beetle optimization algorithm:
taking the prediction precision of the CNN-BILSTM hybrid neural network as an objective function, and defining the value range of super parameters, wherein the super parameters comprise hidden layer dimensions, a maximum training period and an initial learning rate;
randomly generating a corresponding number of dung beetle individuals according to the defined value range, wherein each individual represents a super-parameter combination;
and carrying out iterative optimization on each generation of individuals, calculating the objective function value of each individual, and obtaining the optimal super-parameter combination through the optimal individuals recorded in the iterative process.
From the above description, since the depth of the hidden layer determines the expression capability of the neural network, the deep network can learn more complex and abstract features, so that the fitting capability of the model to data is improved; the maximum training period determines the iterative learning times of the model on the training data, and the increase of the training period is helpful for the model to learn the data more fully, but the too large hidden layer or the too large training period is easy to cause the fitting and the calculation cost is increased. Meanwhile, the proper initial learning rate influences the convergence rate of the model, and too large learning rate may cause oscillation or divergence, too small learning rate causes slow convergence and wastes calculation resources. Therefore, the hidden layer dimension, the maximum training period and the initial learning rate are optimized, the cost of the subsequent prediction time can be reduced, the calculation resources are saved, the model performance of the CNN-BILSTM is improved, and the accuracy of the subsequent situation prediction is improved.
Further, the inputting the dimension reduced processed data into the optimized hybrid neural network, and predicting the situation of the power distribution network comprises:
performing bidirectional propagation through a BILSTM neural network to perform time sequence feature selective screening on the input dimensionality reduction processed data;
after the dimension-reduced processed data are transmitted to a full-connection layer, a time-frequency matrix of power grid information signals is obtained by utilizing time-frequency transformation, deep feature extraction is carried out on the dimension-reduced processed data through a CNN layer, and a plurality of groups of time-frequency feature vectors are obtained;
and randomly discarding a plurality of groups of time-frequency characteristic vectors through the Dropout layer to obtain a power grid situation predicted value.
Further, the obtaining the predicted value of the power grid situation includes:
and carrying out result analysis on the power grid situation predicted value, and predicting the change trend of the power grid running state by combining the historical state of the power distribution network.
According to the description, the change trend of the running state of the power grid is predicted, and the potential risk of the power distribution network is monitored in real time conveniently, so that timely early warning, active prevention and control, risk reduction, toughness and elasticity of the power distribution network are improved, and stable running of the intelligent power distribution network is guaranteed.
Further, the collected field operation data includes: at least five data of intelligent power distribution network equipment state information, power distribution network steady state data information, power distribution network dynamic data information, power distribution network transient fault information, power distribution network operation environment information, electric vehicle charging load data, resident power load data, business power load data, distributed power supply output data, three-phase unbalanced load and meteorological data.
From the above description, the on-site operation data is collected through the virtual collection technology of the power distribution network, so that comprehensive perception of the power distribution network can be realized from the two layers of depth and breadth.
Referring to fig. 2, another embodiment of the present invention provides a situation prediction terminal for a power distribution network based on a hybrid neural network, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the steps of the situation prediction method for the power distribution network based on the hybrid neural network when executing the computer program.
The situation prediction method and the terminal of the power distribution network based on the hybrid neural network are suitable for evaluating the running risk of the power grid in real time and accurately predicting the future change trend of the power grid state, and are described by specific embodiments below:
example 1
Referring to fig. 1, a situation prediction method of a power distribution network based on a hybrid neural network includes the steps of:
s1, acquiring field operation data of the power distribution network by using virtual acquisition.
Specifically, the on-site operation data of the intelligent power grid are acquired through a virtual acquisition technology of the power distribution network, and comprehensive perception of the intelligent power grid is achieved from the two layers of depth and breadth. The collected data specifically comprises: at least five data of intelligent power distribution network equipment state information, power distribution network steady state data information, power distribution network dynamic data information, power distribution network transient fault information, power distribution network operation environment information, electric vehicle charging load data, resident power load data, business power load data, distributed power supply output data, three-phase unbalanced load and meteorological data.
S2, carrying out interpolation processing on missing data in the acquired data set, carrying out denoising processing on the acquired data set, and carrying out feature selection and dimension reduction processing on the interpolated data and the denoised data.
In step S2, various data and information collected into the intelligent distribution network need to be integrated, and relationships between behaviors and objects in the data need to be parsed and extracted, so that preparation is made for situation prediction. The method comprises the following steps:
s21, interpolation processing is carried out on missing data in the collected data set of the field operation data by using cubic spline interpolation so as to avoid information loss.
Specifically, an interpolation step h is calculated i ＝x i+1 -x i (i=1, 2, …, n), n representing the number of data in the data set, and the data node being represented as (x) 1 ,y 1 ),(x 2 ,y 2 ),(x 3 ,y 3 ),…,(x n ,y n )；
Calculating coefficient a of spline curve i ,b i ,c i ,d i ：
In each subinterval x i ≤x≤x i+1 In g i (x)＝a i +b i (x-x i )+c i (x-x i ) 2 +d i (x-x i ) 3 ；
Wherein m is i Representing the secondary differential value g i (x) Interpolation data representing the missing value at i.
S22, removing noise in the data set of the collected field operation data by adopting continuous wavelet transformation.
Specifically, the CWT (Continuous Wavelet Transform ) is adopted to remove noise in the original data, and the denoising principle is based on multi-scale analysis and frequency selectivity of wavelet. And the sparsity of the wavelet coefficients is utilized, the threshold value is adaptively selected according to the quality of data in the data set and the characteristics of noise, and the threshold value is processed through the wavelet coefficients, so that the optimal denoising effect is achieved, and the data quality is improved.
S23, adopting L1 norms to perform feature selection on the interpolated data and the denoised data.
Specifically, the L1 norm refers to the sum of absolute values of the individual elements in the vector. In feature selection, the L1 norm is used as a regularization term that can cause some feature weights in the model to become 0, thereby achieving feature sparsification. By sparsifying the feature weights, the L1 norm feature selection can exclude features which do not contribute to the prediction task, and the generalization capability and the interpretation of the model are improved.
In this embodiment, referring to fig. 3, firstly, data in a dataset including features and target variables is normalized to have similar dimensions;
selecting a proper machine learning model, and using the processed data to fit the machine learning model to obtain the weight of the feature;
and adding the L1 norm as a regularization term into a loss function of the model, and controlling the sparseness degree of the feature weight by adjusting regularization parameters.
According to the feature weights, features with importance are selected, and features with features of 0 are excluded from the feature set.
S24, performing dimension reduction processing on the data, wherein the dimension reduction processing comprises the following steps:
the dimension reduction method is KPCA (Kernel Principal Component Analysis), and is a nonlinear dimension reduction algorithm based on a kernel method, and the nonlinear dimension reduction algorithm is used for mapping high-dimension data to a low-dimension space. The implementation flow is as follows:
computing a kernel matrix between the data:
wherein K (x i ,x j ) Representing data x i And x j The result of the kernel function calculation between them,respectively are provided withRepresenting data x i And x j Feature vectors mapped to a high-dimensional space;
the kernel matrix is centered such that each element subtracts the mean of the kernel matrix.
The core matrix is specifically:
and carrying out eigenvalue decomposition on the centralized kernel matrix to obtain eigenvectors and eigenvalues. The eigenvalues represent the projection direction of the data in the low-dimensional space, and the eigenvalues represent the importance of the corresponding eigenvectors. Eigenvalues and eigenvectors: k (K) c α=λα. Wherein K is c Represents a centralised kernel matrix, alpha represents a eigenvector, and lambda represents an eigenvalue.
Then, according to the magnitude of the characteristic values, selecting the first k largest characteristic values and corresponding characteristic vectors as main components; each sample in the high-dimensional data set is projected through the principal component to obtain a corresponding low-dimensional representation:
y i ＝∑ j a j •K(x i ,x j )；
wherein y is i Representing sample x i Projection in low dimensional space, alpha j Representing the feature vector corresponding to the selected principal component.
And S3, optimizing the hidden layer dimension, the maximum training period and the initial learning rate of the hybrid neural network.
Wherein, DBO optimization algorithm (Dung Beetle Optimizer, dung beetle optimization algorithm) is optimized mainly by simulating the behavior of dung beetles rolling dung balls, and the position change of dung beetles responsible for rolling balls is as follows:
z i (t+1)＝z i (t)+η×k×z i (t-1)+b×Δz；
Δz＝|z i (t)-Z w |；
wherein t represents the current iteration number, z i Representing position information of the ith dung beetle; k is expressed as a deflection coefficient; eta e (0, 1) is oneA random number; b is a natural coefficient; z is Z w Representing the global worst position.
The individual dung beetles search the solution space through scrolling and position updating. When encountering an obstacle, the dung beetles dance to reposition the direction, and the positions of the dung beetles are updated as follows: z i (t+1)＝z i (t)+tan(θ)|z i (t)-z i (t-1) wherein tan (θ) represents a deflection coefficient.
The method comprises the steps of guiding the small dung beetles to forage by establishing an optimal foraging area so as to simulate foraging behaviors of the small dung beetles, wherein the optimal foraging area is defined as follows:
Lb b ＝max(Z b ×(1-R),Lb)
Ub b ＝min(Z b ×(1+R),Ub)；
wherein Lb represents the lower boundary, ub represents the upper boundary, Z b Represents a global optimum position, R represents an inertial weight, r=1-T/T max ，T max Representing the maximum number of iterations of the algorithm, t representing the current number of iterations.
Thus, the position of the small dung beetles is updated as follows: z i (t+1)＝z i (t)+C 1 ×(z i (t)-Lb b )+C 2 ×(z i (t)-Ub b ) Wherein C is 1 Representing random numbers subject to normal distribution; c (C) 2 E (0, 1) is a random vector.
Further, assume Z b The method is the best place for competing for food, and the position update of the dung beetles with theft is described as follows: z i (t+1)＝Z b +S×p×(|z i (t)-Z * |+|z i (t)-Z b I), where p represents a random vector subject to a normal too distribution with a mean of 0 and a variance of 1; s is a constant.
The steps of optimizing hidden layer dimension, maximum training period and initial learning rate by adopting DBO are as follows:
1. and taking the prediction precision of the CNN-BILSTM hybrid neural network as an objective function, and defining the value range of the super parameter, wherein the super parameter comprises the dimension of the hidden layer, the maximum training period and the initial learning rate.
2. Initializing a population, namely randomly generating a corresponding number of dung beetle individuals according to a defined value range, wherein each individual represents a super-parameter combination.
3. And carrying out iterative optimization on each generation of individuals, calculating the objective function value of each individual, and obtaining the optimal super-parameter combination through the optimal individuals recorded in the iterative process.
The depth of the hidden layer determines the expression capacity of the neural network, and the deep network can learn more complex and abstract characteristics, so that the fitting capacity of the model to data is improved. The maximum training period determines the iterative learning times of the model on the training data, and the increase of the training period is helpful for the model to learn the data more fully. However, too large a hidden layer or too large a training period easily results in over fitting, increasing the computational cost. The appropriate initial learning rate affects the convergence rate of the model. Too large a learning rate may cause concussion or divergence, too small a learning rate may cause slow convergence, and waste computing resources.
Compared with manual adjustment or a traditional grid search method, the adoption of the DBO intelligent algorithm for super-parameter optimization can save a great deal of time and calculation resources, is beneficial to improving the model performance of CNN-BILSTM and increases the accuracy of situation prediction.
S4, inputting the dimensionality reduction processed data into the optimized hybrid neural network, and predicting the situation of the power distribution network.
The hybrid neural network in this embodiment is a CNN-BILSTM hybrid neural network model, please refer to FIG. 4, which includes an input layer, a BILSTM layer, a fully connected layer, a Dropout layer and an output layer.
S41, performing bidirectional propagation through a BILSTM neural network to perform time sequence feature selective screening on the input dimensionality reduced processed data.
Specifically, the processed data set (intelligent power grid equipment state information, power grid steady state data information, power grid dynamic data information, power grid transient fault information, power grid operation environment information, electric vehicle charging load data, resident electricity load data, business electricity load data, distributed power supply output data, three-phase unbalanced load and meteorological data) is input, and the input intelligent power grid signals are subjected to time sequence feature selective screening through bi-directional propagation of a BILSTM neural network.
S42, after the dimension-reduced processed data are transmitted to the full-connection layer, a time-frequency matrix of the power grid information signal is obtained through time-frequency transformation, deep feature extraction is carried out on the dimension-reduced processed data through the CNN layer, and a plurality of groups of time-frequency feature vectors are obtained.
Specifically, after the signals are transmitted into the full-connection layer, a brand new smart grid information signal time-frequency matrix is obtained by utilizing time-frequency transformation, deep feature extraction is performed on the smart grid signals through the CNN layer, and a plurality of groups of time-frequency feature vectors are obtained.
S43, randomly discarding a plurality of groups of time-frequency characteristic vectors through the Dropout layer to obtain a power grid situation predicted value.
S44, analyzing the result of the power grid situation predicted value, and predicting the change trend of the power grid running state by combining the historical state of the power distribution network.
Specifically, the result analysis is carried out on the predicted value of the situation of the power grid, the future change trend of the running state of the power grid is predicted by combining the historical state of the power grid, the potential risk of the power grid is monitored in real time, early warning and active prevention and control are carried out, the risk is reduced, the toughness and the elasticity of the power grid are improved, and the stable running of the intelligent power grid is ensured.
Example two
Referring to fig. 2, a situation prediction terminal 1 of a power distribution network based on a hybrid neural network includes a memory 2, a processor 3, and a computer program stored in the memory 2 and capable of running on the processor 3, wherein the processor 3 implements the steps of the situation prediction method of the power distribution network based on the hybrid neural network according to the first embodiment when executing the computer program.
In summary, according to the situation prediction method and the terminal for the power distribution network based on the hybrid neural network, firstly, intelligent power grid data are collected through an intelligent power distribution network virtual collection technology; secondly, carrying out interpolation processing on the missing data in the data set by a cubic spline interpolation method, adopting CWT (Continuous Wavelet Transform) to remove noise in the original data, and carrying out feature selection and feature extraction on the data; adopting DBO (Dung Beetle Optimizer) optimization algorithm to optimize three super parameters of the hidden layer dimension of CNN-BILSTM (Convolutional Neural Network-Bi-direction Long Short-Term Memory), the maximum training period and the initial learning rate, and by optimizing the three specific super parameters, the subsequent prediction time cost can be reduced, and the calculation resource can be saved; and finally, predicting the situation of the intelligent power grid by using the DBO-CNN-BILSTM hybrid neural network model. The invention adopts the DBO-CNN-BILSTM hybrid neural network model, can evaluate the running risk of the intelligent power grid in real time, accurately predicts the future change trend of the intelligent power grid state, and timely reminds the distribution management department of safety regulation.
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 (10)

1. A situation prediction method of a power distribution network based on a hybrid neural network is characterized by comprising the following steps:
the method comprises the steps of collecting field operation data of a power distribution network by using virtual collection;
interpolation processing is carried out on the missing data in the acquired data set, denoising processing is carried out on the acquired data set, and feature selection and dimension reduction processing are carried out on the interpolated data and the denoised data;
optimizing the hidden layer dimension, the maximum training period and the initial learning rate of the hybrid neural network;
and inputting the dimensionality-reduced processed data into the optimized mixed neural network, and predicting the situation of the power distribution network.
2. The situation prediction method of a power distribution network based on a hybrid neural network according to claim 1, wherein the interpolating processing of missing data in the collected data set and denoising processing of the collected data set includes:
and carrying out interpolation processing on the missing data in the data set of the collected field operation data by using cubic spline interpolation, and removing noise in the data set of the collected field operation data by adopting continuous wavelet transformation.
3. The situation prediction method of a power distribution network based on a hybrid neural network according to claim 2, wherein the interpolating the missing data in the data set of the collected field operation data by using a cubic spline interpolation method comprises:
calculating interpolation step h i ＝x i+1 -x i (i=1, 2, …, n), n representing the number of data in the data set, and the data node being represented as (x) 1 ,y 1 ),(x 2 ,y 2 ),(x 3 ,y 3 ),…,(x n ,y n )；
Calculating coefficient a of spline curve i ,b i ,c i ,d i ：
In each subinterval x i ≤x≤x i+1 In g i (x)＝a i +b i (x-x i )+c i (x-x i ) 2 +d i (x-x i ) 3 ；
Wherein m is i Representing the secondary differential value g i (x) Interpolation data representing the missing value at i.
4. The situation prediction method for a power distribution network based on a hybrid neural network according to claim 1, wherein the feature selection of the interpolated data and the denoised data comprises:
and adopting L1 norm to perform feature selection on the interpolated data and the denoised data.
5. A situation prediction method for a power distribution network based on a hybrid neural network according to claim 3, wherein the performing the dimension reduction processing on the data comprises:
computing a kernel matrix between the data:
wherein K (x i ,x j ) Representing data x i And x j The result of the kernel function calculation between them,respectively represent data x i And x j Feature vectors mapped to a high-dimensional space;
centering the core matrix:
and decomposing the characteristic values of the centralized kernel matrix to obtain characteristic vectors and characteristic values, sorting the characteristic values in a descending order according to the sizes of the characteristic values, selecting the front preset characteristic values and the corresponding characteristic vectors after sorting as main components, and projecting each sample in a high-dimensional data set through the main components to obtain a corresponding low-dimensional representation.
6. The situation prediction method of the power distribution network based on the hybrid neural network according to claim 1, wherein optimizing the hidden layer dimension, the maximum training period and the initial learning rate of the hybrid neural network comprises optimizing by adopting a dung beetle optimization algorithm:
taking the prediction precision of the CNN-BILSTM hybrid neural network as an objective function, and defining the value range of super parameters, wherein the super parameters comprise hidden layer dimensions, a maximum training period and an initial learning rate;
randomly generating a corresponding number of dung beetle individuals according to the defined value range, wherein each individual represents a super-parameter combination;
and carrying out iterative optimization on each generation of individuals, calculating the objective function value of each individual, and obtaining the optimal super-parameter combination through the optimal individuals recorded in the iterative process.
7. The situation prediction method of a power distribution network based on a hybrid neural network according to claim 6, wherein the inputting the dimensionality-reduced processing data into the optimized hybrid neural network predicts the situation of the power distribution network, and the method comprises:
performing bidirectional propagation through a BILSTM neural network to perform time sequence feature selective screening on the input dimensionality reduction processed data;
after the dimension-reduced processed data are transmitted to a full-connection layer, a time-frequency matrix of power grid information signals is obtained by utilizing time-frequency transformation, deep feature extraction is carried out on the dimension-reduced processed data through a CNN layer, and a plurality of groups of time-frequency feature vectors are obtained;
and randomly discarding a plurality of groups of time-frequency characteristic vectors through the Dropout layer to obtain a power grid situation predicted value.
8. The situation prediction method of a power distribution network based on a hybrid neural network according to claim 7, wherein the obtaining the predicted value of the situation of the power distribution network comprises:
and carrying out result analysis on the power grid situation predicted value, and predicting the change trend of the power grid running state by combining the historical state of the power distribution network.
9. The situation prediction method for a power distribution network based on a hybrid neural network according to claim 1, wherein the collected field operation data comprises: at least five data of intelligent power distribution network equipment state information, power distribution network steady state data information, power distribution network dynamic data information, power distribution network transient fault information, power distribution network operation environment information, electric vehicle charging load data, resident power load data, business power load data, distributed power supply output data, three-phase unbalanced load and meteorological data.
10. A situation prediction terminal for a power distribution network based on a hybrid neural network, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of a situation prediction method for a power distribution network based on a hybrid neural network according to any one of claims 1 to 9 when the computer program is executed by the processor.
CN202311652456.1A 2023-12-04 2023-12-04 Situation prediction method and terminal for power distribution network based on hybrid neural network Pending CN117709712A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311652456.1A CN117709712A (en) 2023-12-04 2023-12-04 Situation prediction method and terminal for power distribution network based on hybrid neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311652456.1A CN117709712A (en) 2023-12-04 2023-12-04 Situation prediction method and terminal for power distribution network based on hybrid neural network

Publications (1)

Publication Number Publication Date
CN117709712A true CN117709712A (en) 2024-03-15

Family

ID=90143465

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311652456.1A Pending CN117709712A (en) 2023-12-04 2023-12-04 Situation prediction method and terminal for power distribution network based on hybrid neural network

Country Status (1)

Country Link
CN (1) CN117709712A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118171049A (en) * 2024-05-13 2024-06-11 西南交通大学 Big data-based battery management method and system for edge calculation

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118171049A (en) * 2024-05-13 2024-06-11 西南交通大学 Big data-based battery management method and system for edge calculation

Similar Documents

Publication Publication Date Title
Jiang et al. Network intrusion detection based on PSO-XGBoost model
CN110212528B (en) Power distribution network measurement data missing reconstruction method
Dariane et al. Forecasting streamflow by combination of a genetic input selection algorithm and wavelet transforms using ANFIS models
CN110006649A (en) A kind of Method for Bearing Fault Diagnosis based on improvement ant lion algorithm and support vector machines
CN117709712A (en) Situation prediction method and terminal for power distribution network based on hybrid neural network
CN110987436B (en) Bearing fault diagnosis method based on excitation mechanism
Wu Power load forecasts based on hybrid PSO with Gaussian and adaptive mutation and Wv-SVM
CN111460001A (en) Theoretical line loss rate evaluation method and system for power distribution network
CN117390635A (en) Safety monitoring method and system based on big data analysis
CN112149896A (en) Attention mechanism-based mechanical equipment multi-working-condition fault prediction method
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
US20210166131A1 (en) Training spectral inference neural networks using bilevel optimization
Couckuyt et al. Towards efficient multiobjective optimization: multiobjective statistical criterions
CN112488399B (en) Power load prediction method and device
Peng et al. Study of the short-term electric load forecast based on ANFIS
Shaikh et al. Wavelet decomposition impacts on traditional forecasting time series models
Yang et al. Evolutionary algorithms with a coarse-to-fine function smoothing
Bin et al. A short-term power load forecasting method based on eemd-abgru
Kayakuş Forecasting carbon dioxide emissions in Turkey using machine learning methods
CN111563684A (en) Load identification method and device and terminal
CN118393368B (en) Method and device for evaluating battery endurance, storage medium and equipment
Mohammadi et al. New evolutionary methods for optimal design of PID controllers for AVR system
CN111914865A (en) Probability main component analysis method based on random core
Zhang et al. Ensemble optimization approach based on hybrid mode decomposition and intelligent technology for wind power prediction system
CN115965288B (en) IWO (IWO-based optimization BiLSTM) based active power distribution network frequency security assessment method

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