LU500551B1 - Virtual load dominant parameter identification method based on incremental learning - Google Patents
Virtual load dominant parameter identification method based on incremental learning Download PDFInfo
- Publication number
- LU500551B1 LU500551B1 LU500551A LU500551A LU500551B1 LU 500551 B1 LU500551 B1 LU 500551B1 LU 500551 A LU500551 A LU 500551A LU 500551 A LU500551 A LU 500551A LU 500551 B1 LU500551 B1 LU 500551B1
- Authority
- LU
- Luxembourg
- Prior art keywords
- training
- samples
- neural network
- identification
- parameter identification
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 59
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 238000013135 deep learning Methods 0.000 claims abstract description 20
- 238000004088 simulation Methods 0.000 claims abstract description 11
- 125000004122 cyclic group Chemical group 0.000 claims abstract description 7
- 238000004458 analytical method Methods 0.000 claims abstract description 6
- 238000005259 measurement Methods 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 239000002245 particle Substances 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 2
- 238000010801 machine learning Methods 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007429 general method Methods 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
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
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06311—Scheduling, planning or task assignment for a person or group
-
- 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
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A virtual load dominant parameter identification method based on incremental learning, comprising following steps of: (1) randomly selecting dominant parameters from a virtual load model for simulation; (2) establishing a deep learning neural network; (3) performing incremental learning on the neural network; and (4) performing fast online identification and cyclic training. According to the present invention, the feasibility of applying incremental learning to power system analysis is mainly described and the incremental learning is combined with load parameter identification, which improves training efficiency while ensuring identification accuracy, and prevents catastrophic forgetting while maintaining storage overhead, thus providing a new idea for processing training samples in parameter identification, as well as the technical support for online identification of dominant parameters of virtual load models. Based on the idea of continuous training and fast online identification, the convolution neural network is applied to the parameter identification of load models.
Description
DESCRIPTION 750055)
TECHNICAL FIELD The present invention relates to the field of power system load identification, in particular to a virtual load dominant parameter identification method based on incremental learning.
BACKGROUND OF THE PRESENT INVENTION In the actual operation monitoring of a power system, an accurate load model plays a very important role in the safe and stable operation of the power system. Different load models lead to different or even completely opposite results of stability calculation. Therefore, how to establish an accurate load model and obtain accurate model parameters has always been a hot topic for scholars, and has been widely concerned for a long time. There are two main load modeling methods, namely a component-based method and a measurement-based method. The component-based method firstly classifies the loads, and makes component-based analysis on the characteristics of each type of load to obtain the overall characteristics of the loads. The component- based method has disadvantages that the statistical effort is time-consuming and laborious and cannot be performed online. With the rapid development of measurement systems such as WAMS, PMU and SCADA in recent years, the overall measurement and identification method based on measured data has become the mainstream method in the field of load modeling, which is closely related to the application of artificial intelligence and machine learning.
With the development of artificial intelligence and machine learning, many machine learning algorithms have been developed. Most of these algorithms in Batch Learning mode, i.e. it is assumed that all training samples can be obtained at a time before training, the learning process is terminated after these samples are learned, and no new knowledge needs to be learned. However, in 50055 practical application, training samples are usually impossible to be obtained all at once, but are obtained gradually over time, and the information reflected by the samples may also change over time. It takes a lot of time and space to relearn all the data after the arrival of new samples, so the batch learning algorithms no longer meet such demand. In contrast, incremental learning can update knowledge gradually, and can modify and strengthen previous knowledge, so that the updated knowledge can adapt to the newly arrived data without having to re-learn all the data.
SUMMARY OF THE PRESENT INVENTION A technical problem to be solved by the present invention is as follows. Through combination of parameter identification and neural network incremental learning, the bus online measurement waveforms and the waveforms obtained by random value simulation of dominant parameters of a virtual load model are simultaneously brought into training. A virtual load dominant parameter identification process based on incremental learning is proposed to improve the training efficiency while ensuring the identification accuracy, overcomes the contradiction between a large number of increasing measured data and the lack of effective processing methods, and provides a new way for the power grid operators to identify and analyze parameters.
In order to achieve the above objectives of the present invention, the present application provides a virtual load dominant parameter identification method based on incremental learning is provided, including: step 1: randomly selecting dominant parameters from a virtual load model for simulation in a Power System Analysis Synthesis Program (PSASP) to obtain new training samples; step 2: selecting a convolution neural network to establish a deep learning neural network; step 3: performing incremental learning on the samples in the deep learning neural network, where representative samples selected from previous 50055 samples are sent to the deep learning network established in the step 2 and combined with the new training samples generated in the step 1 for training, so as to achieve training the new samples with different batches of the training samples in the training process while constraining errors of the previous samples; and step 4: performing fast online identification and cyclic training: sending effective online measurement waveforms obtained from a measurement module such as PMU (phasor measurement unit) to the deep learning neural network trained in the step 3 for parameter identification, and sending the identified parameters to the PSASP for simulation and evaluation to obtain measurement samples; and sending the measurement samples into the training samples in the step 3 for cyclic training.
Further, the incremental learning method related to network parameter training in the deep learning neural network in the step 3 includes following steps of: selecting the representative samples from the previous samples of the deep learning neural network to form a sample database; combining the selected representative samples with the new training samples generated in the step 1 proportionally and sending the combined samples to the deep learning neural network for further training, so as to achieve training the new samples with different batches of the training samples in the training process while constraining errors of the previous samples.
Further, the cyclic training method in the step 4 includes following steps of: sending the effective online measurement data obtained from the measurement module to the trained deep learning neural network trained in the step 3 for parameter identification, and sending the identified parameters to the PSASP for simulation; comparing the simulated waveforms with the measured waveforms, and evaluating the identification accuracy: If the fitting is accurate, outputting and saving the identification results to the sample database for follow-up training; if the fitting has large error, identifying the parameters 50055 separately by a Particle Swarm Optimization (PSO) algorithm, and saving the results to the sample database for follow-up training, and repeating this cycle.
Beneficial Effects
1. According to the present invention, the feasibility of applying incremental learning to power system analysis is mainly described and the incremental learning is combined with load parameter identification, which improves training efficiency while ensuring identification accuracy, and prevents catastrophic forgetting while maintaining storage overhead, thus providing a new idea for processing training samples in parameter identification.
2. The solution provided in the present invention also provides technical support for online identification of dominant parameters of virtual load models. Based on the idea of continuous training and fast online identification, the convolution neural network is applied to the parameter identification of load models, and dominant parameters of virtual load models are identified online on the big data platform of a power grid accordingly.
BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a flowchart according to the present invention, and FIG. 2 is a structure diagram of virtual loads according to the present invention.
DETAILED DESCRIPTION OF THE PRESENT INVENTION The present invention will be described in detail with reference to the accompanying drawings.
As shown in FIG. 1, the present invention provides a virtual load dominant parameter identification method based on incremental learning, including following steps 1 to 4.
In the step 1, dominant parameters are randomly selected from a virtual load model for simulation: virtual load is the aggregation of multi-source heterogeneous loads, which emphasizes the function and effectiveness of the 50055 loads as a whole to a large power grid, and the loads are aggregated through advanced control, metering, communication and other technologies without changing the access mode of existing loads. The model established by integration of various traditional models and addition of a distributed new energy model is more accurate and detailed and refined than traditional composite load models, and is more conducive to the coordinated and optimal dispatching of the loads of the large power grid (as shown in FIG. 2). A large number of training samples can be obtained by randomly selecting and simulating the dominant parameters in a certain range.
Z and / are the proportions of constant impedance and constant current in a static load model (Zp+/p+Pp=1, the same as Za), Km is the proportion of active power of a motor in the total load, Lr is the induction motor slip, and Koc is the proportion of distributed energy active power. The above parameters are the dominant parameters that have a great influence on the response waveform, and a significant influence on disturbance waveforms when changing. Other parameters (such as internal parameters of a motor) are less sensitive to changes, so IEEE typical values can be adopted. Simulation is performed by an |IEEE-39 bus system in a Power System Analysis Synthesis Program (PSASP), the voltage drops randomly by grounding adjacent lines, and the waveforms U, P and Q are measured at an outlet of a bus to be identified. The loads of the bus are simulated by a virtual load aggregation model to generate a large number of disturbance waveforms for training.
In the step 2, a deep learning neural network is established and a convolution neural network is selected. A standard feedforward neural network consists of an input layer, a hidden layer and an output layer. The input is generally a one-dimensional column vector, and the layers are all fully connected layers, which can be regarded as a directed acyclic graph describing the correlation between functions. The classification and prediction capacity of the deep learning network is enhanced by feature extraction methods such as adding a plurality of convolution and pooling layers before the input layer of the 50055 standard feedforward neural network, and neurons are randomly deleted in the hidden layer to prevent overfitting during training.
In the step 3, incremental learning is performed in the neural network: representative samples selected from previous samples are sent to the deep learning network established in the step 2 and combined with the new training samples generated in the step 1 for training, so as to achieve training the new samples with different batches of the training samples in the training process while constraining errors of the previous samples.
The representative samples are defined to reflect the vast majority of the characteristics of an entire data set by some samples in the data set, thus avoiding reviewing the entire data set every time when reading the data set. The general method herein for selecting the representative samples is to select the samples with the smallest classification error or to take a mean center of each class, corresponding to supervised learning and unsupervised learning, respectively.
In the step 4, fast online identification includes the steps of sending effective online measurement data obtained from a measurement module such as PMU to the neural network trained in the step 3 for parameter identification, and sending the identified results to the PSASP for simulation. Comparing the simulated waveforms with the measured waveforms, and evaluating the identification accuracy: if the fitting is accurate, outputting and saving the identified results to the sample database for follow-up training; if the fitting has large error, identifying the parameters separately by a Particle Swarm Optimization (PSO) algorithm, and saving the results to the sample database for follow-up training, that is, using the results as the training samples to perform the step 3.
Claims (3)
1. A virtual load dominant parameter identification method based on incremental learning, comprising: step 1: randomly selecting dominant parameters from a virtual load model for simulation in a Power System Analysis Synthesis Program (PSASP) to obtain new training samples; step 2: selecting a convolution neural network to establish a deep learning neural network; step 3: performing incremental learning on the samples in the deep learning neural network, where representative samples selected from previous samples are sent to the deep learning network established in the step 2 and combined with the new training samples generated in the step 1 for training, so as to achieve training the new samples with different batches of the training samples in the training process while constraining errors of the previous samples; and step 4: performing fast online identification and cyclic training: sending effective online measurement waveforms obtained from a measurement module such as PMU to the deep learning neural network trained in the step 3 for parameter identification, and sending the identified parameters to the PSASP for simulation and evaluation to obtain measurement samples; and sending the measurement samples into the training samples in the step 3 for cyclic training.
2. The virtual load dominant parameter identification method based on incremental learning according to claim 1, wherein the incremental learning method related to the network parameter training in the deep learning neural network in the step 3 comprises following steps of: selecting the representative samples from the previous samples of the deep learning neural network to form a sample database; combining the selected representative samples with the new training samples generated in the step 1 proportionally and sending the combined samples to the deep learning neural network for further training, so as to achieve training the new 50055 samples with different batches of the training samples in the training process while constraining errors of the previous samples.
3. The virtual load dominant parameter identification method based on incremental learning according to claim 1, wherein the cyclic training method in the step 4 comprises following steps of: sending the effective online measurement data obtained from the measurement module to the trained deep learning neural network trained in the step 3 for parameter identification, and sending the identified parameters to the PSASP for simulation; comparing the simulated waveforms with the measured waveforms, and evaluating the identification accuracy: If the fitting is accurate, outputting and saving the identification results to the sample database for follow-up training; if the fitting has large error, identifying the parameters separately by a Particle Swarm Optimization (PSO) algorithm, and saving the results to the sample database for follow-up training, and repeating this cycle.
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011080582.0A CN112241836B (en) | 2020-10-10 | 2020-10-10 | Virtual load leading parameter identification method based on incremental learning |
Publications (1)
Publication Number | Publication Date |
---|---|
LU500551B1 true LU500551B1 (en) | 2022-02-18 |
Family
ID=74168671
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
LU500551A LU500551B1 (en) | 2020-10-10 | 2021-08-18 | Virtual load dominant parameter identification method based on incremental learning |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN112241836B (en) |
LU (1) | LU500551B1 (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113536674B (en) * | 2021-07-13 | 2023-09-29 | 国网浙江省电力有限公司湖州供电公司 | Line parameter identification method based on BP neural network and improved SCADA data |
CN113850302B (en) * | 2021-09-02 | 2023-08-29 | 杭州海康威视数字技术股份有限公司 | Incremental learning method, device and equipment |
CN114049143B (en) * | 2021-10-29 | 2022-07-22 | 湖南大学 | Node-holiday power load-oriented derivative cluster model prediction method and system |
Family Cites Families (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP5554987B2 (en) * | 2009-12-28 | 2014-07-23 | キヤノン株式会社 | Object identification device and control method thereof |
CN101825869A (en) * | 2010-05-13 | 2010-09-08 | 上海交通大学 | Method for identifying superheater model parameters based on data drive |
CN105809286B (en) * | 2016-03-08 | 2021-08-03 | 南昌工程学院 | Incremental SVR load prediction method based on representative data reconstruction |
CN106446942A (en) * | 2016-09-18 | 2017-02-22 | 兰州交通大学 | Crop disease identification method based on incremental learning |
CN107622308B (en) * | 2017-09-18 | 2020-07-10 | 华中科技大学 | Power generation equipment parameter early warning method based on DBN (database-based network) |
CN107742029A (en) * | 2017-10-19 | 2018-02-27 | 国家电网公司 | Increasing knowledge and magnanimity based on SVMs are super to return load modeling multi-cure-fitting model |
CN109784529A (en) * | 2018-12-06 | 2019-05-21 | 国网甘肃省电力公司金昌供电公司 | A kind of prediction technique and device of electric load |
CN109784748B (en) * | 2019-01-25 | 2022-06-21 | 广东电网有限责任公司 | User electricity consumption behavior identification method and device under market competition mechanism |
US20200265301A1 (en) * | 2019-02-15 | 2020-08-20 | Microsoft Technology Licensing, Llc | Incremental training of machine learning tools |
CN110045606B (en) * | 2019-03-25 | 2021-07-27 | 中南大学 | Increment space-time learning method for online modeling of distributed parameter system |
CN110543906B (en) * | 2019-08-29 | 2023-06-16 | 彭礼烨 | Automatic skin recognition method based on Mask R-CNN model |
CN110782014A (en) * | 2019-10-23 | 2020-02-11 | 新华三信息安全技术有限公司 | Neural network increment learning method and device |
-
2020
- 2020-10-10 CN CN202011080582.0A patent/CN112241836B/en active Active
-
2021
- 2021-08-18 LU LU500551A patent/LU500551B1/en active IP Right Grant
Also Published As
Publication number | Publication date |
---|---|
CN112241836B (en) | 2022-05-20 |
CN112241836A (en) | 2021-01-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
LU500551B1 (en) | Virtual load dominant parameter identification method based on incremental learning | |
CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
Luo et al. | Data-driven short-term voltage stability assessment based on spatial-temporal graph convolutional network | |
Li et al. | An intelligent transient stability assessment framework with continual learning ability | |
CN116245033B (en) | Artificial intelligent driven power system analysis method and intelligent software platform | |
CN111478314B (en) | Transient stability evaluation method for power system | |
CN114006370B (en) | Power system transient stability analysis and evaluation method and system | |
CN114169231A (en) | Method for obtaining distribution line fault classification, positioning and line selection deep learning model based on transfer learning | |
CN112149873A (en) | Low-voltage transformer area line loss reasonable interval prediction method based on deep learning | |
CN116087692B (en) | Distribution network tree line discharge fault identification method, system, terminal and medium | |
CN112821424A (en) | Power system frequency response analysis method based on data-model fusion drive | |
CN116706992A (en) | Self-adaptive power prediction method, device and equipment for distributed photovoltaic cluster | |
CN115358437A (en) | Power supply load prediction method based on convolutional neural network | |
Wang et al. | Distribution network state estimation based on attention-enhanced recurrent neural network pseudo-measurement modeling | |
CN117132132A (en) | Photovoltaic power generation power prediction method based on meteorological data | |
CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
Shi et al. | A fault location method for distribution system based on one-dimensional convolutional neural network | |
CN112232570A (en) | Forward active total electric quantity prediction method and device and readable storage medium | |
CN113949079B (en) | Power distribution station user three-phase unbalance prediction optimization method based on deep learning | |
CN115983714A (en) | Static security assessment method and system for edge graph neural network power system | |
CN116204771A (en) | Power system transient stability key feature selection method, device and product | |
Keyan et al. | Anomaly detection method of distribution network line loss based on hybrid clustering and LSTM | |
Indralaksono et al. | Hierarchical Clustering and Deep Learning for Short-Term Load Forecasting with Influenced Factors | |
Ning | Comparison of several common intelligent fault diagnosis knowledge-based method under nonlinear small sample conditions | |
Xingjia et al. | Hadoop Based Data Mining and Short-Term Power Load Forecasting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
FG | Patent granted |
Effective date: 20220218 |