CN113887794A - Method and device for reactive power optimization of power distribution network - Google Patents
Method and device for reactive power optimization of power distribution network Download PDFInfo
- Publication number
- CN113887794A CN113887794A CN202111144517.4A CN202111144517A CN113887794A CN 113887794 A CN113887794 A CN 113887794A CN 202111144517 A CN202111144517 A CN 202111144517A CN 113887794 A CN113887794 A CN 113887794A
- Authority
- CN
- China
- Prior art keywords
- network
- reactive power
- power distribution
- dbn
- power optimization
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N3/045—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
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/46—Controlling of the sharing of output between the generators, converters, or transformers
- H02J3/50—Controlling the sharing of the out-of-phase component
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- 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
-
- 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)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Power Engineering (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a method and a device for reactive power optimization of a power distribution network, wherein the method comprises the following steps: inputting historical load, topological data and a corresponding historical reactive power optimization strategy of the power distribution network into the improved CNN network for feature extraction; historical load and topological data in the power distribution network data after the characteristic extraction are used as input of the DBN, a corresponding historical reactive power optimization strategy is used as output of the DBN, the DBN is trained, and the training is stopped until a preset iteration number or network loss is unchanged to obtain a reactive power optimization model; inputting the historical load and the topological data acquired in real time into a reactive power optimization model to obtain a real-time reactive power optimization strategy, and performing reactive power optimization on the power distribution network by using the real-time reactive power optimization strategy; the invention has the advantages that: the calculation time of the reactive power optimization process of the power distribution network is short and the precision is high.
Description
Technical Field
The invention relates to the field of reactive power optimization of a power distribution network, in particular to a method and a device for reactive power optimization of the power distribution network.
Background
The integration of large-scale distributed power sources into a traditional power distribution network brings great challenges to the topology, the operation state, the control method and the like of the large-scale distributed power sources. The negative influence on the power distribution network caused by the grid connection of the distributed power supply is solved. The number of electric devices for improving the operation state of the power grid is gradually increased. The reactive power optimization of the power distribution network refers to a process of minimizing voltage fluctuation and active loss of the power distribution network by adjusting active power and reactive power of a load node, a transformer tap transformation ratio, reactive compensation capacity of the load node and the like on the premise of safety and stability of a power system.
The traditional reactive power optimization solving method needs to utilize and simplify a physical model of the power distribution network, and an optimal solution is found through a strategy iteration method. For example, chinese patent publication No. CN110729740A provides a method, an apparatus, a computer device and a readable storage medium for reactive power optimization of a power distribution network, the method includes: the method comprises the steps of obtaining a reactive power optimization model of a power distribution network system, converting a power distribution network system network loss optimization function and a discrete reactive equipment action loss optimization function into a Markov decision process corresponding to the power distribution network reactive power optimization according to the reactive power optimization model, solving the Markov decision process by taking a state value as a target to obtain configuration information of discrete reactive power regulation equipment, and configuring the power distribution network system according to the configuration information to realize the reactive power optimization of the power distribution network system. The large-scale distributed power supply and the intervention of the power equipment increase the complexity of the reactive power optimization model to a certain extent. Not only the calculation time is increased, but also the calculation precision is lower. In order to ensure that the operation of the power distribution network is safer, more economic and more stable, a new reactive power optimization method which is short in calculation time, high in precision and independent of a physical model is urgently needed.
In recent years, with the rapid development of big data technology in various industries, experts in various fields pay more and more attention to the big data technology. The power distribution network covers a large amount of historical load data in the data acquisition process. Because the geographic positions of the load data sources are close, the correlation among the data is strong. This makes it possible to apply data-driven technology to reactive power optimization in a power distribution network. The data driving means that a deep learning technology is utilized to analyze a large amount of data, so that effective data information is obtained and is used for guiding the subsequent reactive power optimization process of the power distribution network. The data driving technology is mainly divided into two categories in the field of reactive power optimization of power distribution networks: a similarity matching algorithm and a classifier training algorithm. The method mainly comprises a random matrix theory, an association rule and the like, and the method mainly comprises a support vector machine, a deep confidence network and a convolution neural network. Although large-dimensional data can be processed by using the random matrix theory and the association rule, the two methods have low data processing speed and general algorithm convergence. The SVM can utilize the CNN to perform reactive power optimization, the network computing speed is high due to the fact that a network structure is not a full connection layer, but a topological structure between nodes is omitted in a traditional CNN structure, so that a part of effective information of a power distribution network is lost, and the computing accuracy is more difficult to improve.
Disclosure of Invention
The invention aims to solve the technical problems of long calculation time and low precision of the reactive power optimization method of the power distribution network in the prior art.
The invention solves the technical problems through the following technical means: a method of reactive power optimization of a power distribution network, the method comprising the steps of:
the method comprises the following steps: inputting historical load, topological data and a corresponding historical reactive power optimization strategy of the power distribution network into the improved CNN network for feature extraction; the improved CNN network comprises 3 convolutional layers, 3 pooling layers and 1 full-connection layer, and is connected in a mode of convolution-pooling-full-connection layer;
step two: historical load and topological data in the power distribution network data after the characteristic extraction are used as input of a DBN (distributed binary network), a corresponding historical reactive power optimization strategy is used as output of the DBN, network parameters are set, the DBN connection weight is continuously updated through a particle swarm optimization algorithm, the DBN is trained, and the training is stopped until a preset iteration number or network loss is unchanged to obtain a reactive power optimization model;
step three: inputting the historical load and the topological data acquired in real time into a reactive power optimization model, obtaining a real-time reactive power optimization strategy through the trained reactive power optimization model, and performing reactive power optimization on the power distribution network by using the real-time reactive power optimization strategy.
The method automatically adjusts the DBN network connection weight through a particle calculation optimization algorithm, uses a historical characteristic index as input, uses a historical reactive power optimization strategy as output, learns the functional relation between the characteristic index and the reactive power optimization strategy through the DBN network to obtain a reactive power optimization model, uses data extracted by the current moment characteristic as input, and obtains a real-time reactive power optimization strategy through the learnt reactive power optimization model.
Further, the first step comprises:
combining the connection node relation matrix of the power distribution network with historical load data of the power distribution network to form a mixed characteristic matrix A;
preprocessing the mixed characteristic matrix by a formula B which is A + I to obtain an adjacent order matrix B, wherein A represents a connection node relation matrix of the power distribution network, and I represents a unit matrix;
for the adjacent order matrix B through the formulaCarrying out normalization to obtain an updated mixed feature matrix, wherein D is a diagonal matrix;
and inputting the updated mixed feature matrix into the improved CNN network for feature extraction.
Further, one convolution layer and one pooling layer in the improved CNN network in the step one are used as single-layer convolutional neural networks, and the convolution kernel size of the convolution layer and the pooling layer in each single-layer convolutional neural network is selected through an error rate minimum trial selection method.
Further, the error rate minimization test selection method comprises:
in the single-layer convolutional neural network, firstly fixing the size of a convolutional kernel to calculate the error rate of the improved CNN network, then continuously adjusting the size of the convolutional kernel, calculating the error rate of the improved CNN network when the size of the convolutional kernel is adjusted each time, stopping adjusting the size of the convolutional kernel when the error rate of the improved CNN network is minimum or reaches a preset error rate value, and taking the size of the currently obtained convolutional kernel as the finally determined size of the convolutional kernel of the single-layer convolutional neural network.
Further, the DBN network comprises a plurality of RBM networks and a BP neural network which are sequentially cascaded.
Further, the continuously updating the DBN network connection weight through the particle swarm optimization algorithm includes:
step 201: setting a plurality of groups of DBN network connection weights, and taking each set of DBN network connection weights as a particle value of one particle;
step 202: setting parameters of the particle swarm, namely the swarm size, the learning factor, the inertia weight and the maximum iteration number of the particles;
step 203: according to the particle value by formulaCalculating a fitness value of the particle, wherein N represents the number of samples, m represents the dimension of the particle, pijRepresenting the reconstructed value, t, of the j-th dimension of the particle in the i-th sampleijRepresenting the actual value of the j dimension particle in the ith sample;
step 204: comparing the fitness value of each particle with the group extreme value P, if the fitness value of the current particle is greater than the group extreme value P, the current particle is the optimal particle, and updating the DBN network by using the DBN network connection weight corresponding to the optimal particle; and if the fitness value of the current particle is not greater than the extreme value P in the group, updating the speed and the position of each particle, updating the particle value, and returning to execute the step 203.
Further, in the second step, after the DBN network connection weight is updated each time, the input and output data set of the DBN network is divided into a test set and a training set for DBN network training, and the ratio of the test set to the training set is 3: 7.
The invention also provides a device for reactive power optimization of the power distribution network, which comprises:
the characteristic extraction module is used for inputting the historical load, the topological data and the corresponding historical reactive power optimization strategy of the power distribution network into the improved CNN network for characteristic extraction; the improved CNN network comprises 3 convolutional layers, 3 pooling layers and 1 full-connection layer, and is connected in a mode of convolution-pooling-full-connection layer;
the model optimization module is used for taking historical load and topological data in the power distribution network data after feature extraction as input of a DBN (distributed binary network), taking a corresponding historical reactive power optimization strategy as output of the DBN, setting network parameters, continuously updating DBN network connection weights through a particle swarm optimization algorithm, training the DBN, and stopping training until preset iteration times or network loss is unchanged to obtain a reactive power optimization model;
and the reactive power optimization module is used for inputting the historical load and the topological data acquired in real time into the reactive power optimization model, obtaining a real-time reactive power optimization strategy through the trained reactive power optimization model, and performing reactive power optimization on the power distribution network by using the real-time reactive power optimization strategy.
Further, the feature extraction module is further configured to:
combining the connection node relation matrix of the power distribution network with historical load data of the power distribution network to form a mixed characteristic matrix A;
preprocessing the mixed characteristic matrix by a formula B which is A + I to obtain an adjacent order matrix B, wherein A represents a connection node relation matrix of the power distribution network, and I represents a unit matrix;
for the adjacent order matrix B through the formulaNormalizing to obtain an updated mixed feature matrix, wherein D isA diagonal matrix;
and inputting the updated mixed feature matrix into the improved CNN network for feature extraction.
Further, one convolution layer and one pooling layer in the improved CNN network in the feature extraction module are used as single-layer convolutional neural networks, and the sizes of convolution kernels of the convolution layer and the pooling layer in each single-layer convolutional neural network are selected through an error rate minimum trial selection method.
Further, the error rate minimization test selection method comprises:
in the single-layer convolutional neural network, firstly fixing the size of a convolutional kernel to calculate the error rate of the improved CNN network, then continuously adjusting the size of the convolutional kernel, calculating the error rate of the improved CNN network when the size of the convolutional kernel is adjusted each time, stopping adjusting the size of the convolutional kernel when the error rate of the improved CNN network is minimum or reaches a preset error rate value, and taking the size of the currently obtained convolutional kernel as the finally determined size of the convolutional kernel of the single-layer convolutional neural network.
Further, the DBN network comprises a plurality of RBM networks and a BP neural network which are sequentially cascaded.
Further, the continuously updating the DBN network connection weight through the particle swarm optimization algorithm includes:
step 201: setting a plurality of groups of DBN network connection weights, and taking each set of DBN network connection weights as a particle value of one particle;
step 202: setting parameters of the particle swarm, namely the swarm size, the learning factor, the inertia weight and the maximum iteration number of the particles;
step 203: according to the particle value by formulaCalculating a fitness value of the particle, wherein N represents the number of samples, m represents the dimension of the particle, pijRepresenting the reconstructed value, t, of the j-th dimension of the particle in the i-th sampleijRepresenting the actual value of the j dimension particle in the ith sample;
step 204: comparing the fitness value of each particle with the group extreme value P, if the fitness value of the current particle is greater than the group extreme value P, the current particle is the optimal particle, and updating the DBN network by using the DBN network connection weight corresponding to the optimal particle; and if the fitness value of the current particle is not greater than the extreme value P in the group, updating the speed and the position of each particle, updating the particle value, and returning to execute the step 203.
Furthermore, after the DBN network connection weight is updated every time in the model optimization module, the input and output data set of the DBN network is divided into a test set and a training set for DBN network training, and the ratio of the test set to the training set is 3: 7.
The invention has the advantages that: the method automatically adjusts the DBN network connection weight through a particle calculation optimization algorithm, uses a historical characteristic index as input, uses a historical reactive power optimization strategy as output, learns the functional relation between the characteristic index and the reactive power optimization strategy through the DBN network to obtain a reactive power optimization model, uses data extracted by the current moment characteristic as input, and obtains a real-time reactive power optimization strategy through the learnt reactive power optimization model.
Drawings
Fig. 1 is a schematic diagram of an improved CNN network structure in a method for reactive power optimization of a power distribution network according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a DBN network structure in a method for reactive power optimization of a power distribution network according to an embodiment of the present invention;
fig. 3 is a mathematical representation of a reactive power optimization model in a method for reactive power optimization of a power distribution network according to an embodiment of the present invention;
fig. 4 is a flowchart of a method for reactive power optimization of a power distribution network according to an embodiment of the present invention;
fig. 5 is a schematic view of a topology structure of an IEEE33 node in the method for power distribution network reactive power optimization disclosed in the embodiment of the present invention;
fig. 6 is a comparison between the method for reactive power optimization of a power distribution network disclosed in the embodiment of the present invention and voltage deviations of PSO, SVM, DBN, and CNN;
fig. 7 is a comparison between the reactive power optimization method for the power distribution network and the network loss of the PSO, the SVM, the DBN, and the CNN according to the embodiment of the present invention;
fig. 8 is a comparison between the method for reactive power optimization of a power distribution network disclosed in the embodiment of the present invention and the comprehensive functions of PSO, SVM, DBN, and CNN.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
As shown in fig. 4, a method for reactive power optimization of a power distribution network includes the following steps:
the method comprises the following steps: inputting historical load, topological data and a corresponding historical reactive power optimization strategy of the power distribution network into the improved CNN network for feature extraction; the improved CNN network comprises 3 convolutional layers, 3 pooling layers and 1 full-connection layer, and is connected in a mode of convolution-pooling-full-connection layer; the specific process is as follows:
combining the connection node relation matrix of the power distribution network with historical load data of the power distribution network to form a mixed characteristic matrix A;
in order to solve the problems, an identity matrix is added to a mixed feature matrix A, namely the mixed feature matrix is preprocessed through a formula B (A + I) to obtain an adjacent order matrix B, wherein A represents a connection node relation matrix of the power distribution network, and I represents the identity matrix;
due to the zero-order matrix andin order to solve the problems, B is normally subjected to standard processing, and an adjacent order matrix B is subjected to formula processingCarrying out normalization to obtain an updated mixed feature matrix, wherein D is a diagonal matrix; at a certain moment, the adjacent matrix has the order of 33 × 33 and the load is 1 × 33, so the new feature matrix can be obtained by multiplying the two matrixes, and because the adjacent matrix is not changed at each moment but the load is changed constantly, when the new feature matrix is calculated, only different load data and the adjacent matrix need to be multiplied to obtain the feature matrix of 1 × 33 at each moment, and the new mixed feature matrix is formed. And calculating the mixed feature matrix at each moment, namely using the mixed feature matrix data at each moment as initial data of the CNN model. And inputting the updated mixed feature matrix into the improved CNN network for feature extraction.
The convolution kernel of a conventional CNN network is usually chosen to be odd. In order to enhance the feature extraction effect, the convolution kernel size cannot be too large, ranging from 1 × 1, 3 × 3, 5 × 5, 7 × 7, 9 × 9. As shown in fig. 1, one convolutional layer and one pooling layer in the improved CNN network are used as single-layer convolutional neural networks, and the convolutional kernel size of the convolutional layer and the pooling layer in each single-layer convolutional neural network is selected by an error rate minimum trial selection method, so as to determine the convolutional kernel range. The error rate minimum trial selection method comprises the following steps: in the single-layer convolutional neural network, firstly fixing the size of a convolutional kernel to calculate the error rate of the improved CNN network, then continuously adjusting the size of the convolutional kernel, calculating the error rate of the improved CNN network when the size of the convolutional kernel is adjusted each time, stopping adjusting the size of the convolutional kernel when the error rate of the improved CNN network is minimum or reaches a preset error rate value, and taking the size of the currently obtained convolutional kernel as the finally determined size of the convolutional kernel of the single-layer convolutional neural network. In this example, step (1): in a single-layer convolutional neural network (C1 layer and S2 layer), the size of a convolutional kernel is fixed, then the batch size is increased in sequence, the size of the convolutional kernel is changed, the steps are repeated, and the optimal number of convolutional kernels and the batch size of C1 and S2 are obtained through experimental results. Step (2): and (3) inputting parameters in the upper network as a C3 layer, repeating the step (1), and obtaining the number of the optimal convolution kernels of C3 and S4 and the number of batch sizes through experimental results. And (3): and (3) inputting parameters in the upper network as a C5 layer, repeating the step (1), and obtaining the number of the optimal convolution kernels of C5 and S6 and the number of batch sizes through experimental results.
Step two: historical load and topological data in the power distribution network data after the characteristic extraction are used as input of a DBN (distributed binary network), a corresponding historical reactive power optimization strategy is used as output of the DBN, network parameters are set, the DBN connection weight is continuously updated through a particle swarm optimization algorithm, the DBN is trained, and the training is stopped until a preset iteration number or network loss is unchanged to obtain a reactive power optimization model; the reactive power optimization of the power distribution network is realized by selecting a proper reactive power regulation mode, so that the voltage fluctuation and the network loss of the power distribution network are minimum. From the aspect of functions, the key point is how to optimize the DBN by using a particle swarm optimization algorithm to enable a functional mapping relation to be established among loads X of each node of the power distribution network, a topological structure A and a power equipment state Y, and when the PSO-DBN model is used for reactive power optimization, the model is used for learning the mapping relation among loads, the topological structure of the nodes and the power equipment control strategy by inputting historical data processed by the improved CNN and the power equipment control strategy output historically. And then, obtaining a real-time power equipment control strategy through the learned mapping relation by using the real-time data processed by the CNN.
As shown in fig. 2, the DBN network includes a plurality of RBM networks and a BP neural network, which are sequentially cascaded. And after the DBN network connection weight is updated every time, dividing an input/output data set of the DBN network into a test set and a training set for DBN network training, wherein the ratio of the test set to the training set is 3: 7.
At the beginning of building the DBN model, the initial weight of the DBN network is obtained through random assignment, if the assignment is improper, a local convergence phenomenon can occur in the DBN model in training, and therefore the initial weight of the DBN network is optimized through a particle swarm optimization. The step of continuously updating the DBN network connection weight through the particle swarm optimization algorithm comprises the following steps:
step 201: setting a plurality of groups of DBN network connection weights, and taking each set of DBN network connection weights as a particle value of one particle;
step 202: setting parameters of the particle swarm, namely the swarm size, the learning factor, the inertia weight and the maximum iteration number of the particles;
step 203: according to the particle value by formulaCalculating a fitness value of the particle, wherein N represents the number of samples, m represents the dimension of the particle, pijRepresenting the reconstructed value, t, of the j-th dimension of the particle in the i-th sampleijRepresenting the actual value of the j dimension particle in the ith sample;
step 204: comparing the fitness value of each particle with the group extreme value P, if the fitness value of the current particle is greater than the group extreme value P, the current particle is the optimal particle, and updating the DBN network by using the DBN network connection weight corresponding to the optimal particle; and if the fitness value of the current particle is not greater than the extreme value P in the group, updating the speed and the position of each particle, updating the particle value, and returning to execute the step 203.
Step three: as shown in fig. 3, historical load and topology data acquired in real time are input into a reactive power optimization model, a real-time reactive power optimization strategy is obtained through the trained reactive power optimization model, and reactive power optimization is performed on the power distribution network by using the real-time reactive power optimization strategy, wherein the PSO-DBN network is a DBN network optimized by a particle swarm optimization, namely, the reactive power optimization model.
The effect of the invention is proved by the following simulation experiment effect: as shown in FIG. 5, the IEEE33 node system is taken as a research object, reactive compensation is added to nodes 22 and 33 of the system in FIG. 5, the transformer tap transformation ratio is adjusted in branches 16 to 17, 19 to 20, 24 to 25 and 26 to 27, the size of the historical data amount is two months, and 1440 sample data sets are measured in each hour.
In order to make the reactive power optimization effect better, the invention needs to determine relevant parameters of the improved CNN network and the DBN network (PSO-DBN) optimized by the particle swarm optimization. In the improved CNN feature extraction part, 1120 groups are used as training sets, and 280 groups are used as test sets. The activation function is RELU. The feature extraction capability of the network parameters is measured according to the error rate minimum trial selection method, and the minimum error rate value of the c1 layer is the minimum when the batch size is 32 and the convolution kernel is 7 x 7, namely the optimal network parameters are obtained. Similarly, layers c3 and c5 are the best network parameters for batch sizes of 16 and 32 and convolution kernel sizes of 5 x 5 and 3 x 3, respectively. And in the parameter selection part of the PSO-DBN, the particle swarm parameters are unchanged, and the network parameters of the DBN are improved.
The results of the method of the present invention are shown in the figures, in comparison to PSO, SVM, DBN and CNN. Fig. 6 to 8 illustrate the voltage deviation, the network loss and the index of the comprehensive function of the five optimization methods respectively, and the method of the invention has the advantages of small voltage deviation, small network loss, small comprehensive function, high explanation convergence speed and short calculation time.
Through the technical scheme, the invention firstly adds node topological structure information into a network input layer by using a power distribution network connection node relation matrix, forms a mixed characteristic matrix by combining power distribution network load data, then changes the traditional CNN structure into a LeNet7 structure and adds the node topological structure information into the network input, compared with the traditional CNN, the structure is added with an implicit layer and a pooling layer, the characteristic extraction capability of the CNN can be enhanced, the DBN network connection weight is automatically adjusted by a particle calculation optimization algorithm, the historical characteristic index is used as the input, the historical reactive power optimization strategy is a function relation between the DBN network learning characteristic index and the reactive power optimization strategy is output to obtain a reactive power optimization model, the data extracted at the current moment is used as the input, and the real-time reactive power optimization strategy is obtained by the learned reactive power optimization model, the model is simple, the model is optimized through training, the calculation time is short, the precision is high, the improved CNN network comprises 3 convolution layers, 3 pooling layers and 1 full-connection layer, the number of layers is more than that of the prior art, the original data are more effectively extracted, and the precision is further improved.
Example 2
Based on embodiment 1, embodiment 2 of the present invention further provides a device for power distribution network reactive power optimization, where the device includes:
the characteristic extraction module is used for inputting the historical load, the topological data and the corresponding historical reactive power optimization strategy of the power distribution network into the improved CNN network for characteristic extraction; the improved CNN network comprises 3 convolutional layers, 3 pooling layers and 1 full-connection layer, and is connected in a mode of convolution-pooling-full-connection layer;
the model optimization module is used for taking historical load and topological data in the power distribution network data after feature extraction as input of a DBN (distributed binary network), taking a corresponding historical reactive power optimization strategy as output of the DBN, setting network parameters, continuously updating DBN network connection weights through a particle swarm optimization algorithm, training the DBN, and stopping training until preset iteration times or network loss is unchanged to obtain a reactive power optimization model;
and the reactive power optimization module is used for inputting the historical load and the topological data acquired in real time into the reactive power optimization model, obtaining a real-time reactive power optimization strategy through the trained reactive power optimization model, and performing reactive power optimization on the power distribution network by using the real-time reactive power optimization strategy.
Specifically, the feature extraction module is further configured to:
combining the connection node relation matrix of the power distribution network with historical load data of the power distribution network to form a mixed characteristic matrix A;
preprocessing the mixed characteristic matrix by a formula B which is A + I to obtain an adjacent order matrix B, wherein a represents a connection node relation matrix of the power distribution network, and I represents a unit matrix;
for the adjacent order matrix B through the formulaCarrying out normalization to obtain an updated mixed feature matrix, wherein D is a diagonal matrix;
and inputting the updated mixed feature matrix into the improved CNN network for feature extraction.
Specifically, one convolution layer and one pooling layer in the improved CNN network in the feature extraction module are used as single-layer convolutional neural networks, and the sizes of convolution kernels of the convolution layer and the pooling layer in each single-layer convolutional neural network are selected through a minimum error rate trial selection method.
More specifically, the error rate minimization test selection method includes:
in the single-layer convolutional neural network, firstly fixing the size of a convolutional kernel to calculate the error rate of the improved CNN network, then continuously adjusting the size of the convolutional kernel, calculating the error rate of the improved CNN network when the size of the convolutional kernel is adjusted each time, stopping adjusting the size of the convolutional kernel when the error rate of the improved CNN network is minimum or reaches a preset error rate value, and taking the size of the currently obtained convolutional kernel as the finally determined size of the convolutional kernel of the single-layer convolutional neural network.
Specifically, the DBN network comprises a plurality of RBM networks and a BP neural network which are sequentially cascaded.
Specifically, the continuously updating the DBN network connection weight through the particle swarm optimization algorithm includes:
step 201: setting a plurality of groups of DBN network connection weights, and taking each set of DBN network connection weights as a particle value of one particle;
step 202: setting parameters of the particle swarm, namely the swarm size, the learning factor, the inertia weight and the maximum iteration number of the particles;
step 203: according to the particle value by formulaCalculating a fitness value of the particle, wherein N represents the number of samples, m represents the dimension of the particle, pijRepresenting the reconstructed value, t, of the j-th dimension of the particle in the i-th sampleijRepresenting the actual value of the j dimension particle in the ith sample;
step 204: comparing the fitness value of each particle with the group extreme value P, if the fitness value of the current particle is greater than the group extreme value P, the current particle is the optimal particle, and updating the DBN network by using the DBN network connection weight corresponding to the optimal particle; and if the fitness value of the current particle is not greater than the extreme value P in the group, updating the speed and the position of each particle, updating the particle value, and returning to execute the step 203.
Specifically, after the DBN network connection weight is updated each time in the model optimization module, the input and output data set of the DBN network is divided into a test set and a training set for DBN network training, and the ratio of the test set to the training set is 3: 7.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for reactive power optimization of a power distribution network is characterized by comprising the following steps:
the method comprises the following steps: inputting historical load, topological data and a corresponding historical reactive power optimization strategy of the power distribution network into the improved CNN network for feature extraction; the improved CNN network comprises 3 convolutional layers, 3 pooling layers and 1 full-connection layer, and is connected in a mode of convolution-pooling-full-connection layer;
step two: historical load and topological data in the power distribution network data after the characteristic extraction are used as input of a DBN (distributed binary network), a corresponding historical reactive power optimization strategy is used as output of the DBN, network parameters are set, the DBN connection weight is continuously updated through a particle swarm optimization algorithm, the DBN is trained, and the training is stopped until a preset iteration number or network loss is unchanged to obtain a reactive power optimization model;
step three: inputting the historical load and the topological data acquired in real time into a reactive power optimization model, obtaining a real-time reactive power optimization strategy through the trained reactive power optimization model, and performing reactive power optimization on the power distribution network by using the real-time reactive power optimization strategy.
2. The method for reactive power optimization of the power distribution network according to claim 1, wherein the first step comprises:
combining the connection node relation matrix of the power distribution network with historical load data of the power distribution network to form a mixed characteristic matrix A;
preprocessing the mixed characteristic matrix by a formula B which is A + I to obtain an adjacent order matrix B, wherein A represents a connection node relation matrix of the power distribution network, and I represents a unit matrix;
for the adjacent order matrix B through the formulaCarrying out normalization to obtain an updated mixed feature matrix, wherein D is a diagonal matrix;
and inputting the updated mixed feature matrix into the improved CNN network for feature extraction.
3. The method for reactive power optimization of the power distribution network according to claim 1, wherein one convolutional layer and one pooling layer in the modified CNN network in the first step are used as single-layer convolutional neural networks, and the convolutional kernel size of the convolutional layer and the pooling layer in each single-layer convolutional neural network is selected through an error rate minimization method.
4. The method for reactive power optimization of the power distribution network according to claim 3, wherein the error rate minimization trial selection method comprises:
in the single-layer convolutional neural network, firstly fixing the size of a convolutional kernel to calculate the error rate of the improved CNN network, then continuously adjusting the size of the convolutional kernel, calculating the error rate of the improved CNN network when the size of the convolutional kernel is adjusted each time, stopping adjusting the size of the convolutional kernel when the error rate of the improved CNN network is minimum or reaches a preset error rate value, and taking the size of the currently obtained convolutional kernel as the finally determined size of the convolutional kernel of the single-layer convolutional neural network.
5. The method for reactive power optimization of the power distribution network according to claim 1, wherein the DBN network comprises a plurality of RBM networks and a BP neural network which are sequentially cascaded.
6. The method for reactive power optimization of a power distribution network according to claim 1, wherein the continuously updating DBN network connection weights through a particle swarm optimization algorithm comprises:
step 201: setting a plurality of groups of DBN network connection weights, and taking each set of DBN network connection weights as a particle value of one particle;
step 202: setting parameters of the particle swarm, namely the swarm size, the learning factor, the inertia weight and the maximum iteration number of the particles;
step 203: according to the particle value by formulaCalculating a fitness value of the particle, wherein N represents the number of samples, m represents the dimension of the particle, pijRepresenting the reconstructed value, t, of the j-th dimension of the particle in the i-th sampleijRepresenting the actual value of the j dimension particle in the ith sample;
step 204: comparing the fitness value of each particle with the group extreme value P, if the fitness value of the current particle is greater than the group extreme value P, the current particle is the optimal particle, and updating the DBN network by using the DBN network connection weight corresponding to the optimal particle; and if the fitness value of the current particle is not greater than the extreme value P in the group, updating the speed and the position of each particle, updating the particle value, and returning to execute the step 203.
7. The method for reactive power optimization of the power distribution network according to claim 1, wherein in the second step, after the DBN network connection weight is updated each time, the input/output data set of the DBN network is divided into a test set and a training set for DBN network training, and a ratio of the test set to the training set is 3: 7.
8. An apparatus for reactive power optimization of a power distribution network, the apparatus comprising:
the characteristic extraction module is used for inputting the historical load, the topological data and the corresponding historical reactive power optimization strategy of the power distribution network into the improved CNN network for characteristic extraction; the improved CNN network comprises 3 convolutional layers, 3 pooling layers and 1 full-connection layer, and is connected in a mode of convolution-pooling-full-connection layer;
the model optimization module is used for taking historical load and topological data in the power distribution network data after feature extraction as input of a DBN (distributed binary network), taking a corresponding historical reactive power optimization strategy as output of the DBN, setting network parameters, continuously updating DBN network connection weights through a particle swarm optimization algorithm, training the DBN, and stopping training until preset iteration times or network loss is unchanged to obtain a reactive power optimization model;
and the reactive power optimization module is used for inputting the historical load and the topological data acquired in real time into the reactive power optimization model, obtaining a real-time reactive power optimization strategy through the trained reactive power optimization model, and performing reactive power optimization on the power distribution network by using the real-time reactive power optimization strategy.
9. The device for reactive power optimization of the power distribution network according to claim 8, wherein the feature extraction module is further configured to:
combining the connection node relation matrix of the power distribution network with historical load data of the power distribution network to form a mixed characteristic matrix A;
preprocessing the mixed characteristic matrix by a formula B which is A + I to obtain an adjacent order matrix B, wherein A represents a connection node relation matrix of the power distribution network, and I represents a unit matrix;
for the adjacent order matrix B through the formulaCarrying out normalization to obtain an updated mixed feature matrix, wherein D is a diagonal matrix;
and inputting the updated mixed feature matrix into the improved CNN network for feature extraction.
10. The device for reactive power optimization of the power distribution network according to claim 8, wherein one convolutional layer and one pooling layer in the improved CNN network in the feature extraction module are used as single-layer convolutional neural networks, and the convolutional kernel size of the convolutional layer and the pooling layer in each single-layer convolutional neural network is selected through an error rate minimum trial selection method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111144517.4A CN113887794A (en) | 2021-09-28 | 2021-09-28 | Method and device for reactive power optimization of power distribution network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111144517.4A CN113887794A (en) | 2021-09-28 | 2021-09-28 | Method and device for reactive power optimization of power distribution network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113887794A true CN113887794A (en) | 2022-01-04 |
Family
ID=79007620
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111144517.4A Pending CN113887794A (en) | 2021-09-28 | 2021-09-28 | Method and device for reactive power optimization of power distribution network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113887794A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116526568A (en) * | 2023-07-03 | 2023-08-01 | 国网北京市电力公司 | Method, system, equipment and medium for optimizing distributed power supply of alternating current/direct current power distribution network |
CN117688886A (en) * | 2024-02-02 | 2024-03-12 | 中国人民解放军陆军装甲兵学院 | Parameter optimization method for compensation circuit of wireless power transmission system |
-
2021
- 2021-09-28 CN CN202111144517.4A patent/CN113887794A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116526568A (en) * | 2023-07-03 | 2023-08-01 | 国网北京市电力公司 | Method, system, equipment and medium for optimizing distributed power supply of alternating current/direct current power distribution network |
CN116526568B (en) * | 2023-07-03 | 2023-09-15 | 国网北京市电力公司 | Method, system, equipment and medium for optimizing distributed power supply of alternating current/direct current power distribution network |
CN117688886A (en) * | 2024-02-02 | 2024-03-12 | 中国人民解放军陆军装甲兵学院 | Parameter optimization method for compensation circuit of wireless power transmission system |
CN117688886B (en) * | 2024-02-02 | 2024-04-30 | 中国人民解放军陆军装甲兵学院 | Parameter optimization method for compensation circuit of wireless power transmission system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110535146B (en) | Electric power system reactive power optimization method based on depth determination strategy gradient reinforcement learning | |
CN111738512A (en) | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model | |
CN105631483A (en) | Method and device for predicting short-term power load | |
CN109523084A (en) | A kind of ultrashort-term wind power prediction method based on pivot analysis and machine learning | |
CN109615146B (en) | Ultra-short-term wind power prediction method based on deep learning | |
CN113887794A (en) | Method and device for reactive power optimization of power distribution network | |
CN104636985A (en) | Method for predicting radio disturbance of electric transmission line by using improved BP (back propagation) neural network | |
CN109412161B (en) | Power system probability load flow calculation method and system | |
CN116031888B (en) | Dynamic load prediction-based power flow optimization method, system and storage medium | |
CN114006370B (en) | Power system transient stability analysis and evaluation method and system | |
CN112418496B (en) | Power distribution station energy storage configuration method based on deep learning | |
CN112287990A (en) | Model optimization method of edge cloud collaborative support vector machine based on online learning | |
CN111355633A (en) | Mobile phone internet traffic prediction method in competition venue based on PSO-DELM algorithm | |
CN113139596A (en) | Optimization algorithm of low-voltage transformer area line loss neural network | |
CN111525587A (en) | Reactive load situation-based power grid reactive voltage control method and system | |
CN117132132A (en) | Photovoltaic power generation power prediction method based on meteorological data | |
CN115861671A (en) | Double-layer self-adaptive clustering method considering load characteristics and adjustable potential | |
CN113422371B (en) | Distributed power supply local voltage control method based on graph convolution neural network | |
Cao et al. | Deep learning-based online small signal stability assessment of power systems with renewable generation | |
CN112836876B (en) | Power distribution network line load prediction method based on deep learning | |
CN116667369B (en) | Distributed photovoltaic voltage control method based on graph convolution neural network | |
CN116722541A (en) | Power system load prediction method and device based on convolutional neural network | |
CN116565877A (en) | Automatic voltage partition control method based on spectral cluster analysis | |
CN115793456A (en) | Lightweight sensitivity-based power distribution network edge side multi-mode self-adaptive control method | |
Chen et al. | Design of FOPID controller based on improved sparrow search algorithm |
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 |