WO2020143253A1 - 一种基于Sparse Autoencoder的电力系统运行方式聚类方法 - Google Patents
一种基于Sparse Autoencoder的电力系统运行方式聚类方法 Download PDFInfo
- Publication number
- WO2020143253A1 WO2020143253A1 PCT/CN2019/108714 CN2019108714W WO2020143253A1 WO 2020143253 A1 WO2020143253 A1 WO 2020143253A1 CN 2019108714 W CN2019108714 W CN 2019108714W WO 2020143253 A1 WO2020143253 A1 WO 2020143253A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- power system
- model
- operation mode
- system operation
- training
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Program-control systems
- G05B19/02—Program-control systems electric
- G05B19/04—Program control other than numerical control, i.e. in sequence controllers or logic controllers
- G05B19/042—Program control other than numerical control, i.e. in sequence controllers or logic controllers using digital processors
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06N3/0495—Quantised networks; Sparse networks; Compressed networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0895—Weakly supervised learning, e.g. semi-supervised or self-supervised learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/20—Pc systems
- G05B2219/26—Pc applications
- G05B2219/2639—Energy management, use maximum of cheap power, keep peak load low
Definitions
- the invention belongs to the technical field of power system safety verification and planning operation, and in particular relates to a clustering method of power system operation mode based on Sparse Autoencoder.
- the present invention proposes a method of extracting feature vectors characterizing the operation mode of the power system using Sparse Autoencoder technology.
- the technical problem to be solved by the present invention is to provide a clustering method of the power system operation mode based on Sparse Autoencoder in view of the above-mentioned shortcomings in the prior art.
- the topology and weight matrix are used for cluster analysis to obtain the number of typical scenes, and the original data at the center of each scene is decoded.
- the relevant data forms an input matrix of n rows and m columns n is a vector and m is the number of samples.
- the relevant data includes data of voltages, voltage amplitudes, active power and reactive power of each node in the power system, and time-series load data of the power system within the study time range.
- ⁇ is the coefficient of the L2 regularization method
- ⁇ is the coefficient of sparse regularization
- set the number of hidden layers to a single layer, that is l 1
- step S202 if the Euclidean distance from the original input data after the restoration of the input data is greater than e, increase the number of iterations and retrain the model; if the training model time is greater than t, the error reaches the range early in the iteration and the number of iterations is reduced , Retrain the model.
- step S204 the matrix is calculated versus The Euclidean distance of ⁇ d is accepted if ⁇ d ⁇ .
- the present invention has at least the following beneficial effects:
- the present invention is a clustering method of power system operation mode based on Sparse Autoencoder.
- Sparse Autoencoder technology is applied to the selection of power system feature vectors, which does not require complex and tedious manual standardization of data.
- the training model can find the input At the same time, it is more important to reduce the dimension of the feature vector, determine the initial number of clustering scenes, and greatly reduce the complexity of clustering time.
- the relevant data reflects the main characteristics of the power system operation. Using the relevant data as input can increase the speed and accuracy of the Sparse Autoencoder training model.
- the initial training parameters are flexibly set, the number of hidden layers and the number of neurons are convenient for training in different situations.
- the accuracy of the model can be improved, and the feature vectors can be accurately extracted, which provides good conditions for cluster analysis.
- the scene clustering contour value is obtained, which is used to judge the pros and cons of the model and modify the model.
- the lowest-level feature vector features obtained by the training are restored, and compared with the input vector to determine the restoration degree and error of the model. If the requirements are met, the model is available.
- the lowest-level feature vector features obtained by training are restored and compared with the input vector. If the error is too large, the parameters are modified to retrain the model.
- the present invention can quickly select and reduce the dimensionality of the feature vectors that characterize the operation mode of the power system, and provide a new idea and method for the selection of the feature vectors of the power system operation mode and the generation of typical operation scenarios. At the same time, it sets a precedent for the application of neural networks in this area.
- Figure 1 is the program flow chart of Sparse Autoencoder
- Figure 2 is a simplified diagram of Sparse Autoencoder's algorithm.
- the neural network tries to learn a h W,b (x) ⁇ The function of x. At this time, reducing the number of neurons will force the neural network to learn the compressed representation of the input data, which realizes the process of data dimensionality reduction.
- this algorithm is more suitable for power systems because it helps to find correlations in the input data.
- the measure of the average output activation of a neuron is defined as:
- this penalty factor will punish those There is a significantly different situation from ⁇ so that the average activity of hidden neurons is kept within a small range.
- the specific form of the penalty factor we will choose the following:
- s 1 is the number of hidden neurons in the hidden layer, and the index j in turn represents each neuron in the hidden layer.
- Regularization is an important method in machine learning to prevent overfitting, because the actual model may not be that complicated, and the model topology and weight matrix learned at the same time may only perform better on the training data. With too many features used, it is easy to fall into overfitting when there are few samples. Then we need to transform it into a simpler model.
- the present invention uses the L2 regularization method.
- L is the number of hidden layers
- n is the number of observations
- k is the number of variables in the training set.
- ⁇ is the coefficient of the L2 regularization method
- ⁇ is the coefficient of sparse regularization, which can be modified separately by the L2WeightRegularization and SparsityRegularization functions.
- the invention provides a clustering method of power system operation mode based on Sparse Autoencoder to obtain relevant data in the power system, such as: node voltage, voltage amplitude, node load, generator active and reactive power output in the power system; Then set the training parameters, hide the number of layers and neurons, train the relevant model, and extract the topology and weight matrix of the model, and then perform cluster analysis; finally get the number of typical scenes, and decode to get each scene Raw data from the center.
- the method of the present invention can be used to quickly select and reduce the dimensionality of the feature vectors that characterize the operation mode of the power system, and provide a new idea and method for the selection of the feature vectors of the power system operation mode and the generation of typical operation scenarios. .
- Roughly filter the data of the power system operation for example: obtain the data of the voltage of each node in the system, the active power and reactive power of the generator of each node, and the time-series load data of the system in the research time range. These data constitute n-dimensional, that is, n-row vectors. At the same time, the number of samples is m, which forms an input matrix of n rows and m columns
- the initial input is shown in Table 1. There are a total of 30,000 sample data, each with 53 feature vectors as
- Groups a, b, and c respectively represent data obtained from three different load levels of the IEEE14 node system as input sets. will As input, perform the operation in 2);
- the clustering time and the dimension of the feature vectors participating in the clustering are almost linear, that is, the higher the dimension of the feature vector, the longer the clustering time. It reflects the classification of typical scenes using Sparse Autoencoder. When the clustering effect is almost unchanged, when the feature vector is reduced, it greatly reduces the time consumed, which meets the speed of the power system in the calculation .
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Automation & Control Theory (AREA)
- Human Resources & Organizations (AREA)
- Tourism & Hospitality (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Marketing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Strategic Management (AREA)
- Evolutionary Biology (AREA)
- General Business, Economics & Management (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
一种基于Sparse Autoencoder的电力系统运行方式聚类方法,获取电力系统中的相关数据,然后设置训练参数,隐藏层层数和神经元个数,对相关数据进行Autoencoder模型训练,同时提取出模型的拓扑结构和权值矩阵,进行聚类分析,得出典型场景数目,解码得出各场景中心的原始数据。本方法可以对表征电力系统运行方式的特征向量进行快速选取和降维,为电力系统运行方式特征向量的选取并生成典型运行场景提供一种新思路和方法。同时为神经网络在这一方面的应用开创了一个先例。
Description
本发明属于电力系统安全校验、规划运行技术领域,具体涉及一种基于Sparse Autoencoder的电力系统运行方式聚类方法。
电力系统中使用典型运行方式对于电网的安全运行进行校验有着举足轻重的作用。在规划时期考虑到典型的运行方式并以此进行电力系统的运行校验,可以最大程度的防止电压越限、过负荷等事故的发生,保证电力系统对负荷和用户的持续供电能力。但是随着新能源的不断接入,电力系统的运行随机性大幅度提高,导致运行方式的特征也更加复杂,如何提取运行方式的特征向量生成典型场景变得尤为困难。然而利用传统的PCA方法无法准确地提取特征向量,时间复杂度过高,实用性也大幅度降低。
因此,为保证可靠地提取出表征电力系统运行方式的特征向量以进行典型场景分析,选取合理的特征向量提取方式是亟需认真考虑的问题。
针对上述问题,本发明提出了一种利用Sparse Autoencoder技术提取表征电力系统运行方式的特征向量的方法。
发明内容
本发明所要解决的技术问题在于针对上述现有技术中的不足,提供一种基于Sparse Autoencoder的电力系统运行方式聚类方法。
本发明采用以下技术方案:
一种基于Sparse Autoencoder的电力系统运行方式聚类方法,获取电力系统中的相关数据,然后设置训练参数,隐藏层层数和神经元个数,对相关数据进行Autoencoder模型训练,同时 提取出模型的拓扑结构和权值矩阵,进行聚类分析,得出典型场景数目,解码得出各场景中心的原始数据。
进一步的,相关数据包括电力系统内各个节点电压、电压幅值、各个节点的发电机有功功率和无功功率的数据以及电力系统在研究时间范围内的时序负荷数据。
具体的,设置训练参数,隐藏层层数和神经元个数如下:
设置相关参数为α,η与最大迭代次数为初始化训练参数,α是L2正则化方法的系数,η是稀疏正则化的系数;设置隐藏层层数为单层,即l=1;设置第l隐藏层神经元个数,即最终特征向量维数h
l=2。
具体的,对相关数据进行Autoencoder模型训练步骤具体如下:
S202、输入可接受的误差e以及训练时间t进行可视化训练,观察误差和训练过程;
S203、提取最底层特征向量features
l,并对features
l进行聚类分析;
S205、得出所需结果,循环结束。
进一步的,步骤S202中,若还原输入数据后与原始输入数据的欧氏距离大于e,增加迭代次数,重新训练模型;若训练模型时间大于t,即在迭代早期误差达到范围,减小迭代次数,重新训练模型。
进一步的,步骤S204中,若Φ
d>ε,若l>1,返回l=l-1,重新训练模型;否则,返回h=h-1,重新训练模型。
与现有技术相比,本发明至少具有以下有益效果:
本发明一种基于Sparse Autoencoder的电力系统运行方式聚类方法,将Sparse Autoencoder技术应用于电力系统特征向量的选取上,不需要复杂繁琐的人工标准化数据的过程,通过训练模型可以找到输入量之间的相关性,同时更重要的是可以降低特征向量的维数,确定聚类的初始场景数,同时大大减少了聚类时间上的复杂度。
进一步的,相关数据反应了电力系统运行的主要特征,将相关数据作为输入可以增加Sparse Autoencoder训练模型的速度和精度。
进一步的,根据不同电力系统模型聚类精度的要求,灵活设置初始的训练参数,隐藏层层数和神经元个数,便于针对不同的情况进行训练。
进一步的,通过进行Autoencoder模型训练,可以提高该模型的精度,准确提取出特征向量,为聚类分析提供了良好的条件。
进一步的,通过对Autoencoder模型训练得出的最底层特征向量features,得出场景聚类轮廓值,用于判断模型的优劣以及修改模型。
进一步的,将训练所得最底层特征向量features进行还原,与输入向量进行比较,判断模型的还原度和误差,若满足要求则该模型可用。
进一步的,将训练所得最底层特征向量features进行还原,与输入向量进行比较,若误差过大,则修改参数重新训练模型。
综上所述,本发明可以对表征电力系统运行方式的特征向量进行快速选取和降维,为电力系统运行方式特征向量的选取并生成典型运行场景提供一种新思路和方法。同时为神经网络在这一方面的应用开创了一个先例。
下面通过附图和实施例,对本发明的技术方案做进一步的详细描述。
图1为Sparse Autoencoder的程序流程图;
图2为Sparse Autoencoder的算法简图。
由于Sparse Autoencoder技术可以避免复杂的电力系统数据标准化,训练出的特征向量作聚类分析误差小并且可以在解码后较好地还原电力系统原始数据,具有及其优良的特性,因此选择该技术。
Sparse Autoencoder是一种无监督学习算法,它使用反向传播算法,并让目标值等于输入值,比如y
(i)=x
(i),该神经网络尝试学习一个h
W,b(x)≈x的函数。此时,当减少神经元的个数会强迫该神经网络去学习输入数据的压缩表示,这就实现了数据的降维过程。同时,由于这一算法有利于发现输入数据中的相关性,就更适用于电力系统。
A、稀疏性的定义:
神经元的平均输出激活测度定义为:
其中,ρ是稀疏性参数,通常是一个接近于0的较小的值(比如ρ=0.03)。同时为了实 现这一限制,我们将会在我们的优化目标函数中加入一个额外的惩罚因子,而这一惩罚因子将惩罚那些
和ρ有显著不同的情况从而使得隐藏神经元的平均活跃度保持在较小范围内。惩罚因子的具体形式有很多种合理的选择,我们将会选择以下这一种:
s
1是隐藏层中隐藏神经元的数量,而索引j依次代表隐藏层中的每一个神经元。
B、L2正则化方法:
正则化是机器学习中一个防止过拟合的重要手段,因为实际的模型可能并没有那么复杂,同时学习的模型拓扑以及学到的权值矩阵只是在训练数据上表现的结果可能会比较好。所用的特征过多,样本较少时就很容易陷入过拟合。那么我们就需要把它转化成为更为简单的模型。
本发明使用了L2正则化方法。
定义以下公式:
L是隐藏层的层数,n是观测值的数量,k是训练集中变量的个数。
C、成本函数:
α是L2正则化方法的系数,η是稀疏正则化的系数,可以通过L2WeightRegularization和SparsityRegularization函数去分别修改。
本发明提供了一种基于Sparse Autoencoder的电力系统运行方式聚类方法,获取电力系统中的相关数据,如:电力系统中节点电压、电压幅值、节点负荷、发电机有功和无功出力等;然后设置训练参数,隐藏层层数和神经元个数,训练相关模型,同时提取出模型的拓扑结构 和权值矩阵,然后进行聚类分析;最后得出典型场景数目,并解码得出各场景中心的原始数据。利用本发明方法可以对表征电力系统运行方式的特征向量进行快速选取和降维,为电力系统运行方式特征向量的选取并生成典型运行场景提供一种新思路和方法。。
请参阅图1和图2,本发明一种基于Sparse Autoencoder的电力系统运行方式聚类方法的步骤如下:
S1、数据简单初始化;
粗略进行电力系统运行的数据筛选,例如:获取该系统内各个节点电压、各个节点的发电机有功功率和无功功率的数据以及系统在研究时间范围内的时序负荷数据。这些数据构成n维,即n行向量。同时样本数量共有m个,即形成n行m列输入矩阵,记为
设置相关参数α,η与最大迭代次数,α是L2正则化方法的系数,η是稀疏正则化的系数,即初始化训练参数;第l隐藏层神经元个数,即最终特征向量维数,记为h
l=2;隐藏层层数,默认为单层,记为l=1;
S202、可视化训练过程,观察误差和训练过程,输入可接受的误差e以及训练时间t进行可视化训练,若还原输入数据后与原始输入数据的欧氏距离大于e,增加迭代次数,重新训练模型;若训练模型时间大于t,即在迭代早期误差可达到范围,减小迭代次数,重新训练模型;
S203、提取最底层特征向量,记为features
l,并对features
l进行聚类分析;
选取K-means方法进行聚类,聚类中心个数设为k,初始值设置为k=1,计算轮廓值大小,记作
不断给k=k+1,计算轮廓值大小,记作
当k=h时,退出循环;得出最大的轮 廓值
得出k值,即为典型场景个数;若认为最大轮廓值
小于0.85,当h
l<h
l-1返回设置神经元个数,h
l=h
l+1,重新训练模型;否则,返回设置隐藏层层数,l=l+1,重新训练模型;
若Φ
d>ε:
若l>1,返回l=l-1,重新训练模型;
否则,返回h=h-1,重新训练模型;
S205、得出所需结果,循环结束。
S3、提取出模型拓扑以及学到的权值矩阵,根据需要分析变量的相关性。
提取S2中的最佳k值,即为典型场景个数,并提取出对应的场景中心原始数。
为使本发明实施例的目的、技术方案和优点更加清楚,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本发明一部分实施例,而不是全部的实施例。通常在此处附图中的描述和所示的本发明实施例的组件可以通过各种不同的配置来布置和设计。因此,以下对在附图中提供的本发明的实施例的详细描述并非旨在限制要求保护的本发明的范围,而是仅仅表示本发明的选定实施例。基于本发明中的实施例,本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
下面结合附图和运用IEEE-14节点系统例对本发明做详细说明。
表一输入数据
2、进行模型训练:设置最大迭代次数为1000,α=0.01,η=4;设置初始h
1=2以及l=1等按照2)中方法不断进行循环找出最佳结果;
3、提取出模型拓扑以及学到的权值矩阵,根据需要分析变量的相关性。提取2)中的最 佳k值,即为典型场景个数,并提取出对应的场景中心原始数据。
将计算所得轮廓值展示如下表:
表二轮廓值计算值
由表2可知当典型场景数目为三类时,计算所得的轮廓值最大为0.96左右,得出在该输入数据进行训练的情况下,最佳的聚类水平应该分为三类。其聚类结果符合预期随着负荷水平分类的三种情况,具有极显著的特性。
同时在训练的场景数目不变的情况下,聚类的时间和参与聚类的特征向量的维数几乎成线性关系,即特征向量维数越高,聚类的时间越长。体现了使用Sparse Autoencoder对于典型场景的分类,在聚类效果几乎不变的情况下,当对特征向量进行降维时,其极大减少消耗的时间,这满足了电力系统在计算时的快速性。同时由结果可见,若电网的规模更大即节点数,特征向量维数更高,通过Sparse Autoencoder降低特征向量维数将对于聚类效果的提升有着更加显著的效果,对实际计算有着极大的帮助。
以上内容仅为说明本发明的技术思想,不能以此限定本发明的保护范围,凡是按照本发 明提出的技术思想,在技术方案基础上所做的任何改动,均落入本发明权利要求书的保护范围之内。
Claims (10)
- 一种基于Sparse Autoencoder的电力系统运行方式聚类方法,其特征在于,获取电力系统中的相关数据,然后设置训练参数,隐藏层层数和神经元个数,对相关数据进行Autoencoder模型训练,同时提取出模型的拓扑结构和权值矩阵,进行聚类分析,得出典型场景数目,解码得出各场景中心的原始数据。
- 根据权利要求1或2所述的基于Sparse Autoencoder的电力系统运行方式聚类方法,其特征在于,相关数据包括电力系统内各个节点电压、电压幅值、各个节点的发电机有功功率和无功功率的数据以及电力系统在研究时间范围内的时序负荷数据。
- 根据权利要求1所述的基于Sparse Autoencoder的电力系统运行方式聚类方法,其特征在于,设置训练参数,隐藏层层数和神经元个数如下:设置相关参数为α,η与最大迭代次数为初始化训练参数,α是L2正则化方法的系数,η是稀疏正则化的系数;设置隐藏层层数为单层,即l=1;设置第l隐藏层神经元个数,即最终特征向量维数h l=2。
- 根据权利要求5所述的基于Sparse Autoencoder的电力系统运行方式聚类方法,其特征在于,步骤S202中,若还原输入数据后与原始输入数据的欧氏距离大于e,增加迭代次数,重新训练模型;若训练模型时间大于t,即在迭代早期误差达到范围,减小迭代次数,重新训练模型。
- 根据权利要求5所述的基于Sparse Autoencoder的电力系统运行方式聚类方法,其特征在于,步骤S204中,若Φ d>ε,若l>1,返回l=l-1,重新训练模型;否则,返回h=h-1,重新训练模型。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US17/368,864 US20210334658A1 (en) | 2019-01-08 | 2021-07-07 | Method for performing clustering on power system operation modes based on sparse autoencoder |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910016263.4A CN109711483B (zh) | 2019-01-08 | 2019-01-08 | 一种基于Sparse Autoencoder的电力系统运行方式聚类方法 |
| CN201910016263.4 | 2019-01-08 |
Related Child Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| US17/368,864 Continuation US20210334658A1 (en) | 2019-01-08 | 2021-07-07 | Method for performing clustering on power system operation modes based on sparse autoencoder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020143253A1 true WO2020143253A1 (zh) | 2020-07-16 |
Family
ID=66261049
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/108714 Ceased WO2020143253A1 (zh) | 2019-01-08 | 2019-09-27 | 一种基于Sparse Autoencoder的电力系统运行方式聚类方法 |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US20210334658A1 (zh) |
| CN (1) | CN109711483B (zh) |
| WO (1) | WO2020143253A1 (zh) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113964827A (zh) * | 2021-10-27 | 2022-01-21 | 深圳供电局有限公司 | 基于馈线组特征参数聚类的中压配电网接线模式识别方法 |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711483B (zh) * | 2019-01-08 | 2020-10-27 | 西安交通大学 | 一种基于Sparse Autoencoder的电力系统运行方式聚类方法 |
| CN110990562B (zh) * | 2019-10-29 | 2022-08-26 | 新智认知数字科技股份有限公司 | 警情分类方法及其系统 |
| US12511546B2 (en) * | 2019-10-30 | 2025-12-30 | Royal Bank Of Canada | System and method for machine learning architecture with differential privacy |
| CN111369168B (zh) * | 2020-03-18 | 2022-07-05 | 武汉大学 | 一种适应电网多种调控运行场景的关联特征选择方法 |
| CN111667069B (zh) * | 2020-06-10 | 2023-08-04 | 中国工商银行股份有限公司 | 预训练模型压缩方法、装置和电子设备 |
| CN113704641B (zh) * | 2021-08-27 | 2023-12-12 | 中南大学 | 一种基于拓扑分析的时空大数据潜在结构分析方法 |
| CN115618258B (zh) * | 2022-12-16 | 2023-06-27 | 中国电力科学研究院有限公司 | 一种电力系统规划关键运行方式提取方法及系统 |
| CN116975143B (zh) * | 2023-08-14 | 2025-12-19 | 国网江苏省电力有限公司经济技术研究院 | 一种综合能源系统的典型运行方式提取方法和装置 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106447039A (zh) * | 2016-09-28 | 2017-02-22 | 西安交通大学 | 基于自编码神经网络的无监督特征提取方法 |
| CN108491859A (zh) * | 2018-02-11 | 2018-09-04 | 郭静秋 | 基于自动编码机的驾驶行为异质性特征的识别方法 |
| CN108846410A (zh) * | 2018-05-02 | 2018-11-20 | 湘潭大学 | 基于稀疏自动编码深度神经网络的电能质量扰动分类方法 |
| CN108985330A (zh) * | 2018-06-13 | 2018-12-11 | 华中科技大学 | 一种自编码网络及其训练方法、异常用电检测方法和系统 |
| CN109711483A (zh) * | 2019-01-08 | 2019-05-03 | 西安交通大学 | 一种基于Sparse Autoencoder的电力系统运行方式聚类方法 |
Family Cites Families (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8386249B2 (en) * | 2009-12-11 | 2013-02-26 | International Business Machines Corporation | Compressing feature space transforms |
| US8825565B2 (en) * | 2011-08-25 | 2014-09-02 | Numenta, Inc. | Assessing performance in a spatial and temporal memory system |
| US8756181B2 (en) * | 2011-11-28 | 2014-06-17 | Eaton Corporation | System and method employing a self-organizing map load feature database to identify electric load types of different electric loads |
| CN104904199B (zh) * | 2013-01-11 | 2017-06-06 | 寰发股份有限公司 | 深度查找表的编解码方法及装置 |
| US10073421B2 (en) * | 2015-11-17 | 2018-09-11 | Rockwell Automation Technologies, Inc. | Predictive monitoring and diagnostics systems and methods |
| CN105426839A (zh) * | 2015-11-18 | 2016-03-23 | 清华大学 | 基于稀疏自动编码器的电力系统过电压分类方法 |
| US10776712B2 (en) * | 2015-12-02 | 2020-09-15 | Preferred Networks, Inc. | Generative machine learning systems for drug design |
| CN105550744A (zh) * | 2015-12-06 | 2016-05-04 | 北京工业大学 | 一种基于迭代的神经网络聚类方法 |
| US20170213134A1 (en) * | 2016-01-27 | 2017-07-27 | The Regents Of The University Of California | Sparse and efficient neuromorphic population coding |
| CN106156792A (zh) * | 2016-06-24 | 2016-11-23 | 中国电力科学研究院 | 一种基于台区电气特征参数的低压台区聚类方法 |
| US10248713B2 (en) * | 2016-11-30 | 2019-04-02 | Business Objects Software Ltd. | Time series analysis using a clustering based symbolic representation |
| US20180262525A1 (en) * | 2017-03-09 | 2018-09-13 | General Electric Company | Multi-modal, multi-disciplinary feature discovery to detect cyber threats in electric power grid |
| CN107292531B (zh) * | 2017-07-11 | 2021-01-19 | 华南理工大学 | 一种基于bp神经网络和聚类分析法的公交车“两率”检查方法 |
| CN108229087A (zh) * | 2017-09-30 | 2018-06-29 | 国网上海市电力公司 | 一种低压台区典型场景的析构方法 |
| EP3698268A4 (en) * | 2017-11-22 | 2021-02-17 | Zhejiang Dahua Technology Co., Ltd. | Methods and systems for face recognition |
| US10917415B2 (en) * | 2018-01-10 | 2021-02-09 | Intel Corporation | Machine learning-based determination of program code characteristics |
| CN108459585B (zh) * | 2018-04-09 | 2021-06-01 | 东南大学 | 基于稀疏局部嵌入深度卷积网络的电站风机故障诊断方法 |
| WO2019219198A1 (en) * | 2018-05-17 | 2019-11-21 | Huawei Technologies Co., Ltd. | Device and method for clustering of input-data |
| US10887329B2 (en) * | 2018-07-26 | 2021-01-05 | A10 Networks, Inc. | Cluster-based precision mitigation of network attacks |
| US10834106B2 (en) * | 2018-10-03 | 2020-11-10 | At&T Intellectual Property I, L.P. | Network security event detection via normalized distance based clustering |
| US11470101B2 (en) * | 2018-10-03 | 2022-10-11 | At&T Intellectual Property I, L.P. | Unsupervised encoder-decoder neural network security event detection |
| US11573879B2 (en) * | 2018-10-22 | 2023-02-07 | General Electric Company | Active asset monitoring |
| US10834114B2 (en) * | 2018-12-13 | 2020-11-10 | At&T Intellectual Property I, L.P. | Multi-tiered server architecture to mitigate malicious traffic |
-
2019
- 2019-01-08 CN CN201910016263.4A patent/CN109711483B/zh active Active
- 2019-09-27 WO PCT/CN2019/108714 patent/WO2020143253A1/zh not_active Ceased
-
2021
- 2021-07-07 US US17/368,864 patent/US20210334658A1/en not_active Abandoned
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106447039A (zh) * | 2016-09-28 | 2017-02-22 | 西安交通大学 | 基于自编码神经网络的无监督特征提取方法 |
| CN108491859A (zh) * | 2018-02-11 | 2018-09-04 | 郭静秋 | 基于自动编码机的驾驶行为异质性特征的识别方法 |
| CN108846410A (zh) * | 2018-05-02 | 2018-11-20 | 湘潭大学 | 基于稀疏自动编码深度神经网络的电能质量扰动分类方法 |
| CN108985330A (zh) * | 2018-06-13 | 2018-12-11 | 华中科技大学 | 一种自编码网络及其训练方法、异常用电检测方法和系统 |
| CN109711483A (zh) * | 2019-01-08 | 2019-05-03 | 西安交通大学 | 一种基于Sparse Autoencoder的电力系统运行方式聚类方法 |
Non-Patent Citations (1)
| Title |
|---|
| FU, XIAO ET AL.: "An Optimized Sparse Auto-encoder Network Based on Feature Clustering", CHINESE JOURNAL OF ELECTRONICS, vol. 46, no. 5, 31 May 2018 (2018-05-31), pages 3, XP009521979, ISSN: 2075-5597 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113964827A (zh) * | 2021-10-27 | 2022-01-21 | 深圳供电局有限公司 | 基于馈线组特征参数聚类的中压配电网接线模式识别方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109711483A (zh) | 2019-05-03 |
| CN109711483B (zh) | 2020-10-27 |
| US20210334658A1 (en) | 2021-10-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020143253A1 (zh) | 一种基于Sparse Autoencoder的电力系统运行方式聚类方法 | |
| US11823490B2 (en) | Non-linear latent to latent model for multi-attribute face editing | |
| CN116311454B (zh) | 一种多线索关联学习的面部表情识别方法及系统 | |
| CN109829541A (zh) | 基于学习自动机的深度神经网络增量式训练方法及系统 | |
| CN116050621B (zh) | 一种集成提升模式的多头自注意力海上风电超短时功率预测方法 | |
| CN111507311A (zh) | 一种基于多模态特征融合深度网络的视频人物识别方法 | |
| CN112417289A (zh) | 一种基于深度聚类的资讯信息智能推荐方法 | |
| CN109840595B (zh) | 一种基于群体学习行为特征的知识追踪方法 | |
| CN113535953A (zh) | 一种基于元学习的少样本分类方法 | |
| CN115659254B (zh) | 一种双模态特征融合的配电网电能质量扰动分析方法 | |
| CN112988548A (zh) | 一种基于降噪算法的改进Elman神经网络的预测方法 | |
| CN116090635B (zh) | 一种气象驱动的新能源发电功率预测方法 | |
| CN111275168A (zh) | 基于卷积全连接的双向门控循环单元的空气质量预测方法 | |
| CN119089156A (zh) | 基于lnm大数值模型的业务预测方法、装置及设备 | |
| CN116960978A (zh) | 基于风速-功率组合分解重构的海上风电功率预测方法 | |
| CN113890109A (zh) | 具有时间空间相关性的多风电场功率日场景生成方法 | |
| CN112949610A (zh) | 一种基于降噪算法的改进Elman神经网络的预测方法 | |
| CN119357605A (zh) | 基于时空图神经网络的风电场风速预测方法、介质及设备 | |
| CN111104975A (zh) | 一种基于广度学习的信用评估模型 | |
| CN109960732A (zh) | 一种基于鲁棒监督的深度离散哈希跨模态检索方法及系统 | |
| CN120706994A (zh) | 一种在线学习早期预警方法、装置、设备及介质 | |
| CN116526582B (zh) | 基于人工智能联合驱动的电力机组组合调度方法与系统 | |
| CN120048112A (zh) | 一种基于自注意力-时间卷积网络-生成对抗网络的交通流混合预测模型 | |
| Zhang et al. | Compressing knowledge graph embedding with relational graph auto-encoder | |
| Wang et al. | A convolutional neural network image classification based on extreme learning machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19908698 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19908698 Country of ref document: EP Kind code of ref document: A1 |

