CN110212551B - Micro-grid reactive power automatic control method based on convolutional neural network - Google Patents
Micro-grid reactive power automatic control method based on convolutional neural network Download PDFInfo
- Publication number
- CN110212551B CN110212551B CN201910564883.1A CN201910564883A CN110212551B CN 110212551 B CN110212551 B CN 110212551B CN 201910564883 A CN201910564883 A CN 201910564883A CN 110212551 B CN110212551 B CN 110212551B
- Authority
- CN
- China
- Prior art keywords
- neural network
- reactive power
- convolutional neural
- micro
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/18—Arrangements for adjusting, eliminating or compensating reactive power in networks
-
- 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
- Y02E40/30—Reactive power compensation
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Feedback Control In General (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a convolutional neural network-based microgrid reactive power automatic control method, which uses SCADA to collect real-time operation data of a microgrid system and generate two-dimensional power matrix data; calculating the optimal reactive power of the reactive power device corresponding to the two-dimensional power matrix data by utilizing the optimal power flow, and taking the optimal reactive power as a tag value; the convolutional neural network model is trained to enable it to determine the optimal reactive power for each reactive device based on the system operational data. The invention establishes the convolutional neural network model and carries out model training by utilizing the characteristics of two-dimensional convolutional operation sparse interaction, weight sharing and equal-change representation, realizes automatic feature extraction on the running state of the micro-grid, thereby determining the optimal reactive power of each reactive device, and simultaneously taking into account the voltage deviation and the grid loss during the running of the micro-grid, thereby having good economy and safety.
Description
Technical Field
The invention relates to the field of reactive power automatic control of power systems, in particular to a micro-grid reactive power automatic control method based on a convolutional neural network.
Background
Along with the gradual maturing of the distributed power supply technology and the continuous promotion of commercial projects, a micro-grid system is greatly developed and built as an important technical platform for the efficient management and flexible control of the distributed power supply. However, the fluctuation, intermittence and uncontrollability of wind and light output make the running state of the micro-grid system complex and changeable, and the accurate prediction is difficult. Therefore, as one of means for guaranteeing the stable and economic operation of the micro-grid system, the reactive power automatic control strategy becomes one of the current research hotspots.
At present, the research of the reactive power control method of the micro-grid is concentrated on the traditional control method, and the traditional control method comprises a centralized control mode and a distributed coordination control mode. The former refers to that a dispatching center uniformly controls the tap position of a main transformer of each transformer substation and each reactive compensation device according to the integral operation working condition of the system, and the current common method is a 9-area diagram method. And each reactive power device of the reactive power device independently adjusts the running state of the reactive power device according to the local voltage, so that the voltage quality is improved. However, the control strategies focus on the safety and the electric energy quality of the operation of the micro-grid, namely, the voltage deviation is ensured to be within a reasonable range, and the grid loss is not considered. In addition, a reactive power control strategy based on the optimal power flow is adopted, namely, a multi-objective function considering network loss and voltage deviation is established, and the optimal reactive power of the reactive power device is solved by utilizing an intelligent optimization algorithm according to the running state of the system. However, the intelligent optimization algorithm has long solving time, is easy to sink into local optimum, is not suitable for reactive real-time control, and is mainly used for daily scheduling and real-time scheduling. In consideration of the fact that the convolutional neural network method does not need to build a complex mathematical model aiming at a data set, a calculation result can be obtained rapidly based on input data, the calculation speed is high, and the accuracy is high.
Disclosure of Invention
The invention aims to provide a microgrid reactive power control method based on a convolutional neural network technology, which can effectively reduce the microgrid loss and voltage deviation. The power of each branch and the power of each node acquired by the SCADA are used as input, the modeling and optimizing processes of the system are avoided, and the optimal reactive output of each reactive device is directly obtained. The method comprises the following steps:
step 1, acquiring real-time operation data of a micro-grid system by using SCADA, converting the real-time operation data into a two-dimensional power matrix, and dividing the real-time operation data into a training set and a testing set;
step 2, calculating output values corresponding to the two-dimensional power matrixes based on the optimal power flow, namely the optimal reactive power value of the reactive power device;
step 3, carrying out Gaussian normalization on the two-dimensional power matrix; converting the optimal reactive power value of each reactive power device into a percentage value of the relative self capacity;
and 4, constructing a convolution neural network regression model comprising an input layer, a convolution layer, a pooling layer, a batch standardization layer, a full connection layer and a sigmod layer, and using the convolution neural network regression model for micro-grid reactive power automatic control, wherein the input layer performs data normalization processing on an input two-dimensional power matrix.
Step 5, training a model by using the training set, and evaluating the model effect based on the test set;
and 6, performing reactive power automatic control on the micro-grid system by using the trained convolutional neural network model.
Further, the data amount of the training set and the test set is 4:1.
further, the two-dimensional power matrix is input data of the convolutional neural network.
Further, the output of the convolutional neural network is the reactive power percentage value of each reactive device.
Further, in the convolutional neural network model, 9 layers are added, the convolution kernel of the convolution layer 1 is 1×1, and the other convolution kernels are 3×3.
Further, the structural order of each convolution block in the convolution neural network model is batch normalization, nonlinear activation and convolution except for the convolution layer 1.
Further, the fully connected layer of the convolutional neural network is 3 layers, and the sigmod layer is the last layer of the convolutional neural network.
Further, the sigmod layer activation function is sigmod, the last full-connection layer has no activation function, and other layer activation functions are relu.
Drawings
Fig. 1 is a system structure diagram of a micro-grid reactive power automatic control method based on a convolutional neural network;
fig. 2 is a structural diagram of a convolutional neural network used in a microgrid reactive power automatic control method based on the convolutional neural network;
FIG. 3 is a block diagram of the convolutional neural network of FIG. 1;
fig. 4 is a flowchart of a micro-grid reactive power automatic control method based on a convolutional neural network.
Detailed Description
The present invention will be described in detail with reference to the following examples and drawings.
The invention provides a micro-grid reactive power automatic control method based on a convolutional neural network, which is shown in a figure 1, wherein data such as voltage, current and the like acquired by a scada system are transmitted to a reactive control center through an input information flow, the reactive control center calculates the optimal reactive power of a reactive device by using a convolutional neural network, and an instruction is sent out through an output control flow to adjust the reactive device. The control model establishment comprises the following steps:
step 1: data acquisition
The input of the convolutional neural network is real-time operation data of the micro-grid system, and the real-time operation data comprise active/reactive power of each branch and active/reactive power of each node, and the real-time operation data are represented by a two-dimensional power matrix, as shown in a formula (1).
Wherein:
wherein n is the number of nodes of the micro-grid system; when i+.j, P ij 、Q ij Active power and reactive power are transmitted for node i to node j. When i=j, P ii And Q ii The active power and the reactive power of the node i are respectively, if the active power and the reactive power are positive, the output power is represented, and otherwise, the injection power is represented.
Therefore, the SCADA is utilized to collect operation data of the micro-grid at each moment and convert the operation data into a two-dimensional power matrix. All data were taken as 4:1 is divided into a training set and a test set.
Step 2: data tag generation
The training type of the convolutional neural network model is supervised learning, namely, an optimal model is obtained through training of the existing training samples (namely, known data and corresponding output thereof). Therefore, model training cannot be completed only by two-dimensional power matrix data, and output values corresponding to the data are required to be marked. The output value corresponding to each two-dimensional power matrix data in the invention is the optimal reactive power value of the reactive device, and the expression form is shown in the formula (3).
Wherein m is the number of reactive devices; y is i Reactive power for the ith reactive device.
The SCADA only can collect power data of each branch and each node when the micro-grid operates, and a two-dimensional power matrix is generated to serve as input data of the neural network. The corresponding marked value, namely the optimal reactive power of the reactive power device, needs to be subjected to optimal power flow solving. Therefore, an optimal power flow model is required to be established, and the optimal reactive power of the reactive power device corresponding to each two-dimensional power matrix is determined, as shown in a formula (4):
wherein P is loss Is the network loss of the micro-network; u (U) N Rated voltage for the micro-grid system; u (U) i For the actual voltage at node i lambda 1 、λ 2 Is a weight coefficient.
And (3) carrying out optimal power flow solving on the formula (4) by using an intelligent optimization algorithm, and obtaining the output Y corresponding to each two-dimensional power matrix, namely the optimal reactive power of each reactive power device.
Step 3 data preprocessing
Since the power levels of the branches/nodes are different, the differences between the branches/nodes tend to be large. Thus, if the neural network is directly input, some smaller power values are ignored. The two-dimensional power matrix of the input dataset needs to be normalized with a gaussian distribution as shown in equation (5).
Wherein mu is the average value of the two-dimensional power matrix; delta is the standard deviation of the two-dimensional power matrix.
Because the capacities of the reactive devices are different, if the output of the convolutional neural network is the optimal reactive power value of the reactive device, the magnitude of the loss function of the neural network is influenced by the capacity of the reactive device during training, and the effective evaluation is difficult. For this purpose, the reactive power percentage value of each reactive device is designed and output as shown in a formula (6).
Wherein y is i An actual optimal reactive power value for reactive device i; y is i,N The capacity of the reactive device i;the optimal reactive power relative to capacity percentage value for reactive device i.
Step 4: building convolutional neural networks
The convolutional neural network model used by the invention comprises a convolutional layer, a pooling layer, a full connection layer and a sigmod layer, and the specific structure is shown in fig. 2 and fig. 3. The number of the convolution layers is 9, wherein the convolution kernel size of the 1 st layer is 1 multiplied by 1, the other layers are 3 multiplied by 3, the convolution step length of each layer is 1, and the number of the characteristic layers of each convolution layer is 64, 128, 256, 512 and 512 respectively. To eliminate the overfitting, the convolution layer introduces batch normalization (batch normalization). In addition to convolution layer 1, other convolution layers, batch normalization and nonlinear activation functions, as shown in fig. 2, constitute a convolution block, with the order of the components being batch normalization, nonlinear activation and convolution (BN-RELU-CONV) in sequence. The total connection layer is 3 layers, the node number of each layer is 2048, 4096 and n (n is the number of reactive devices of the micro-grid), the active functions of other layers except the last total connection layer and the sigma mod layer are relu, the active function of the sigma mod layer is sigma mod, and the last total connection layer has no active function.
Defining a loss function of the convolutional neural network as (7):
wherein Y is i ' is the optimal reactive power percentage value of the ith reactive power device output by the neural network model;the optimal reactive power percentage value for the actual ith reactive device.
Step five: training the established neural network model and applying the neural network model to a micro-grid system
The training process is performed by using the training set and the test set, as shown in fig. 1 and fig. 2, and comprises the following steps:
(1) inputting the preprocessed training two-dimensional power matrix into a convolutional neural network, and convolving by using a 1 multiplied by 1 convolution kernel to generate 64 characteristic layers;
(2) the output value of the convolution layer 1 is input into a convolution block 1, and convolution batch standardization, nonlinear activation and convolution calculation are carried out on the output value; nonlinear activation here employs a ReLU-type excitation function. The output value of the convolution block 1 is input to the convolution block 2 in the same manner, and a correlation operation is performed.
(3) The output value of the convolution block 2 is maximally pooled, and the pooled data is input to the convolution block 3, and correlation operation is performed.
(4) The data is passed and correlated in sequence according to fig. 1 until a convolution block 8.
(5) The output of the convolution block 8 is input to the fully-connected layer after being maximally pooled, and is output to the sigmod layer through the three fully-connected layers. The second full-connection layer is a result of performing full-connection after performing nonlinear activation on the first full-connection layer and then performing Dropout; the third full-link layer is not activated by relu non-linearisation.
Dropout is a regularization method that is specifically implemented by setting 0 with 30% probability for each output data of the second fully connected layer, so that it no longer plays any role in the forward operation or backward feedback process. The sigmod layer outputs a percentage value of the optimal reactive power of each reactive device relative to the capacity of each reactive device, and the percentage value is marked as Y';
(6) outputting the result Y according to the sigmod layer i ' and corresponding actual output dataCalculating a loss function;
(7) according to the loss function, reversely adjusting the weight parameters of the convolutional neural network by utilizing an Adam algorithm, and repeating the steps 1 to 6 until the evaluation result of the test set meets the requirement, and outputting a trained convolutional neural network model;
and 6, based on a trained convolutional neural network model, acquiring micro-grid operation data by utilizing a scada system, and generating a control instruction to control the output reactive power of each reactive power device.
The present invention is not limited to the preferred embodiments, and any changes or substitutions that would be apparent to one skilled in the art within the scope of the present invention are intended to be included in the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.
Claims (8)
1. A micro-grid reactive power automatic control method based on a convolutional neural network is characterized in that: the method comprises the following steps:
step 1, acquiring real-time operation data of a micro-grid system by using SCADA, converting the real-time operation data into a two-dimensional power matrix, and dividing the real-time operation data into a training set and a testing set;
step 2, calculating output values corresponding to the two-dimensional power matrixes based on the established optimal power flow model, namely the optimal reactive power value of the reactive power device; the optimal power flow model is as follows:
wherein P is loss Is the network loss of the micro-network; u (U) N Rated voltage for the micro-grid system; u (U) i For the actual voltage at node i lambda 1 、λ 2 Is a weight coefficient;
step 3, carrying out Gaussian normalization on the two-dimensional power matrix; converting the optimal reactive power value of each reactive power device into a percentage value of the relative self capacity;
step 4, constructing a convolution neural network regression model comprising an input layer, a convolution layer, a pooling layer, a batch standardization layer, a full connection layer and a sigmod layer, which is used for micro-grid reactive power automatic control, wherein the input layer carries out data normalization processing on an input two-dimensional power matrix;
step 5, training a model by using the training set, and evaluating the model effect based on the test set;
and 6, performing reactive power automatic control on the micro-grid system by using the trained convolutional neural network model.
2. The automatic control method for micro-grid reactive power based on convolutional neural network according to claim 1, wherein the method comprises the following steps: the data volume of the training set and the test set is 4:1.
3. the automatic control method for micro-grid reactive power based on convolutional neural network according to claim 1, wherein the method comprises the following steps: the two-dimensional power matrix is input data of the convolutional neural network.
4. The automatic control method for micro-grid reactive power based on convolutional neural network according to claim 1, wherein the method comprises the following steps: and the output of the convolutional neural network is the reactive power percentage value of each reactive device.
5. The automatic control method for micro-grid reactive power based on convolutional neural network according to claim 1, wherein the method comprises the following steps: the convolution neural network model has 9 layers, wherein the convolution kernel of the convolution layer 1 is 1 multiplied by 1, and the other convolution kernels are 3 multiplied by 3.
6. The automatic control method for micro-grid reactive power based on convolutional neural network according to claim 1, wherein the method comprises the following steps: the structural order of each convolution block in the convolution neural network model is batch standardization, nonlinear activation and convolution except the convolution layer 1.
7. The automatic control method for micro-grid reactive power based on convolutional neural network according to claim 1, wherein the method comprises the following steps: the full-connection layer of the convolutional neural network is 3 layers, and the sigmod layer is the last layer of the convolutional neural network.
8. The automatic control method for micro-grid reactive power based on convolutional neural network according to claim 1, wherein the method comprises the following steps: the sigmod layer activation function is sigmod, the last full-connection layer has no activation function, and other layer activation functions are relu.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910564883.1A CN110212551B (en) | 2019-06-27 | 2019-06-27 | Micro-grid reactive power automatic control method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910564883.1A CN110212551B (en) | 2019-06-27 | 2019-06-27 | Micro-grid reactive power automatic control method based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110212551A CN110212551A (en) | 2019-09-06 |
| CN110212551B true CN110212551B (en) | 2023-07-21 |
Family
ID=67794890
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910564883.1A Active CN110212551B (en) | 2019-06-27 | 2019-06-27 | Micro-grid reactive power automatic control method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110212551B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112787334A (en) * | 2019-11-08 | 2021-05-11 | 国网辽宁省电力有限公司 | Method and system for rapidly controlling reactive power of battery energy storage power station |
| CN112865118B (en) * | 2021-01-19 | 2022-09-02 | 河海大学 | Deep learning model generation method for power grid dynamic reactive power reserve demand calculation |
| CN113300379B (en) * | 2021-05-08 | 2022-04-29 | 武汉大学 | Electric power system reactive voltage control method and system based on deep learning |
| CN116937579B (en) * | 2023-09-19 | 2023-12-01 | 太原理工大学 | Wind power interval prediction considering space-time correlation and interpretable method thereof |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103151797A (en) * | 2013-03-04 | 2013-06-12 | 上海电力学院 | Multi-objective dispatching model-based microgrid energy control method under grid-connected operation mode |
| CN106655207A (en) * | 2017-03-21 | 2017-05-10 | 国网山东省电力公司枣庄供电公司 | Power distribution network reactive power optimization system and method based on multi-data analysis |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8756047B2 (en) * | 2010-09-27 | 2014-06-17 | Sureshchandra B Patel | Method of artificial nueral network loadflow computation for electrical power system |
| WO2015113637A1 (en) * | 2014-02-03 | 2015-08-06 | Green Power Technologies, S.L. | System and method for the distributed control and management of a microgrid |
| CN108832619A (en) * | 2018-05-29 | 2018-11-16 | 北京交通大学 | Transient stability evaluation in power system method based on convolutional neural networks |
| CN109033702A (en) * | 2018-08-23 | 2018-12-18 | 国网内蒙古东部电力有限公司电力科学研究院 | A kind of Transient Voltage Stability in Electric Power System appraisal procedure based on convolutional neural networks CNN |
| CN109784480B (en) * | 2019-01-17 | 2022-11-18 | 武汉大学 | Power system state estimation method based on convolutional neural network |
-
2019
- 2019-06-27 CN CN201910564883.1A patent/CN110212551B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103151797A (en) * | 2013-03-04 | 2013-06-12 | 上海电力学院 | Multi-objective dispatching model-based microgrid energy control method under grid-connected operation mode |
| CN106655207A (en) * | 2017-03-21 | 2017-05-10 | 国网山东省电力公司枣庄供电公司 | Power distribution network reactive power optimization system and method based on multi-data analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110212551A (en) | 2019-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110212551B (en) | Micro-grid reactive power automatic control method based on convolutional neural network | |
| Ke et al. | Short-term electrical load forecasting method based on stacked auto-encoding and GRU neural network | |
| Jiang et al. | Research and application of an innovative combined model based on a modified optimization algorithm for wind speed forecasting | |
| Capizzi et al. | Recurrent neural network-based control strategy for battery energy storage in generation systems with intermittent renewable energy sources | |
| CN108306303A (en) | A kind of consideration load growth and new energy are contributed random voltage stability assessment method | |
| CN104732300B (en) | A kind of neutral net wind power short term prediction method theoretical based on Fuzzy divide | |
| CN110380444B (en) | Capacity planning method for distributed wind power orderly access to power grid under multiple scenes based on variable structure Copula | |
| CN106600136B (en) | Power section out-of-limit control efficiency evaluation method | |
| CN109523084A (en) | A kind of ultrashort-term wind power prediction method based on pivot analysis and machine learning | |
| CN106529719A (en) | Method of predicting wind power of wind speed fusion based on particle swarm optimization algorithm | |
| CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| CN109829560B (en) | Renewable energy power generation cluster access planning method for power distribution network | |
| CN113300380B (en) | Load curve segmentation-based power distribution network reactive power optimization compensation method | |
| CN112508279B (en) | Regional distributed photovoltaic prediction method and system based on spatial correlation | |
| Khan et al. | Day ahead load forecasting for IESCO using artificial neural network and bagged regression tree | |
| CN104037761A (en) | AGC power multi-objective random optimization distribution method | |
| CN112952807A (en) | Multi-objective optimization scheduling method considering wind power uncertainty and demand response | |
| Panapakidis et al. | A hybrid ANN/GA/ANFIS model for very short-term PV power forecasting | |
| CN115189416A (en) | Power generation system control method and system based on day-ahead electricity price grading prediction model | |
| CN109767353A (en) | A kind of photovoltaic power generation power prediction method based on probability-distribution function | |
| CN112836876A (en) | Power distribution network line load prediction method based on deep learning | |
| CN105207255B (en) | A kind of power system peak regulation computational methods suitable for wind power output | |
| CN109586309B (en) | Power distribution network reactive power optimization method based on big data free entropy theory and scene matching | |
| CN110826776B (en) | Initial solution optimization method based on dynamic programming in distribution network line transformation relation identification | |
| Xia et al. | SCG and LM improved BP neural network load forecasting and programming network parameter settings and data preprocessing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |