CN112016684A - Electric power terminal fingerprint identification method of deep parallel flexible transmission network - Google Patents

Electric power terminal fingerprint identification method of deep parallel flexible transmission network Download PDF

Info

Publication number
CN112016684A
CN112016684A CN202010780676.2A CN202010780676A CN112016684A CN 112016684 A CN112016684 A CN 112016684A CN 202010780676 A CN202010780676 A CN 202010780676A CN 112016684 A CN112016684 A CN 112016684A
Authority
CN
China
Prior art keywords
power terminal
deep
transmission network
parallel
flexible transmission
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.)
Granted
Application number
CN202010780676.2A
Other languages
Chinese (zh)
Other versions
CN112016684B (en
Inventor
殷林飞
马晨骁
罗仕逵
高放
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangxi University
Original Assignee
Guangxi University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangxi University filed Critical Guangxi University
Priority to CN202010780676.2A priority Critical patent/CN112016684B/en
Publication of CN112016684A publication Critical patent/CN112016684A/en
Application granted granted Critical
Publication of CN112016684B publication Critical patent/CN112016684B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
    • G01R19/04—Measuring peak values or amplitude or envelope of AC or of pulses
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
    • G01R23/16—Spectrum analysis; Fourier analysis
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R25/00—Arrangements for measuring phase angle between a voltage and a current or between voltages or currents
    • 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

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Power Engineering (AREA)
  • Image Analysis (AREA)

Abstract

本发明提供一种深度并行柔性发射网络的电力终端指纹识别方法，该方法利用深度并行柔性发射网络模型对电力终端设备运行时的电压和电流波形信息进行学习，形成电力终端指纹与设备名称的关联模型。所提方法是基于柔性发射网络和深度学习结合的深度柔性发射网络模型；且在训练时，所提方法使用不同的深度柔性发射网络参数进行并行学习。在电力系统中新接入电力终端时，所提方法根据其电力终端指纹完成对电力终端的识别。所提方法能对接入电力系统的电力终端进行识别，为电力系统安全运行与控制和负荷预测提供了参考，有效提高电力系统智能化水平。The invention provides a power terminal fingerprint identification method of a deep parallel flexible transmission network. The method uses a deep parallel flexible transmission network model to learn the voltage and current waveform information of the power terminal equipment during operation, and forms an association between the power terminal fingerprint and the device name. Model. The proposed method is a deep flexible transmitting network model based on the combination of flexible transmitting network and deep learning; and during training, the proposed method uses different deep flexible transmitting network parameters for parallel learning. When a power terminal is newly connected to the power system, the proposed method completes the identification of the power terminal according to its power terminal fingerprint. The proposed method can identify the power terminals connected to the power system, provide a reference for the safe operation and control of the power system and load forecasting, and effectively improve the intelligence level of the power system.

Description

一种深度并行柔性发射网络的电力终端指纹识别方法A power terminal fingerprint identification method for deep parallel flexible transmission network

技术领域technical field

本发明属于电力系统智能化与控制领域，涉及一种并行的使用柔性发射网络的深度学习模型对电力终端指纹进行学习并识别的方法，适用于电力系统的负荷识别和智能化水平的提高。The invention belongs to the field of power system intelligence and control, and relates to a parallel method for learning and identifying power terminal fingerprints by using a deep learning model of a flexible transmission network, which is suitable for load identification of power systems and improvement of the level of intelligence.

背景技术Background technique

随着计算机科学的迅速发展，电力系统的智能化水平也在不断提高。现今的智能电网技术已经渗透到了发电、输电、配电、变电和用电的各个环节。基于智能电网技术，电力系统的运行效率和稳定性都得到了提高。在当今人工智能领域快速发展的背景下，电网智能化水平的提高的一个方向是与人工智能的结合。但人工智能模型的训练需要大规模的数据进行支撑，而当今电网的数据源往往是电网设备，因此，将电力终端设备加入到数据来源当中将有效的提升电网运行数据的丰富程度，为电力系统智能化水平的提高奠定基础。电力终端数据中的一个重要数据是接入电网的电力终端的类型、名称。随着智能化电表等终端监测设备的大规模应用，获取电网终端用电数据已经不存在问题。因此，研究有效的电力终端指纹识别方法对这些用电数据进行识别，对了解用户用电特性、提升电网智能化水平、对发电计划进行调整具有重要的指导意义。With the rapid development of computer science, the level of intelligence of power systems is also constantly improving. Today's smart grid technology has penetrated into all aspects of power generation, transmission, distribution, transformation and consumption. Based on smart grid technology, the operation efficiency and stability of the power system have been improved. In the context of the rapid development of the field of artificial intelligence today, one direction of improving the level of intelligence in the power grid is the combination with artificial intelligence. However, the training of artificial intelligence models requires large-scale data to support, and the data source of today's power grid is often power grid equipment. Therefore, adding power terminal equipment to the data source will effectively improve the richness of power grid operation data, which is a good source of power system operation data. The improvement of the level of intelligence lays the foundation. An important data in the power terminal data is the type and name of the power terminal connected to the power grid. With the large-scale application of terminal monitoring equipment such as smart meters, it is no longer a problem to obtain the terminal power consumption data of the power grid. Therefore, researching effective power terminal fingerprint identification methods to identify these power consumption data has important guiding significance for understanding the user's power consumption characteristics, improving the level of power grid intelligence, and adjusting the power generation plan.

发明内容SUMMARY OF THE INVENTION

本发明提出一种深度并行柔性发射网络的电力终端指纹识别方法，本方法使用并行的柔性发射网络的深度学习模型对电力终端指纹进行学习并识别，包括如下步骤：The present invention provides a power terminal fingerprint identification method of a deep parallel flexible transmission network. The method uses the deep learning model of the parallel flexible transmission network to learn and identify the fingerprint of the power terminal, including the following steps:

用电数据获取步骤：采集各个电力终端在单独运行时处于不同工作状态下的电压和电流的用电数据；The step of acquiring power consumption data: collecting the power consumption data of the voltage and current of each power terminal in different working states when running alone;

数据预处理步骤：用电数据获取步骤中得到的电压和电流波形进行傅立叶变换形成4n矩阵作为电力终端指纹特征；Data preprocessing step: Fourier transform is performed on the voltage and current waveforms obtained in the electrical data acquisition step to form a 4n matrix as the fingerprint feature of the power terminal;

训练步骤：目的是建立深度并行柔性发射网络模型，建立电压、电流傅立叶变换矩阵模型与电力终端名称对应的数据集对该模型进行并行训练；对电力终端指纹数据进行训练时，使用基于柔性发射网络和深度学习结合的深度柔性发射网络模型，使用的柔性发射网络神经元层数不小于三层。且在训练时，使用不同的深度柔性发射网络参数进行并行学习，并行模型数不少于三个。所述柔性发射网络神经元框架包括两个前馈通道参数(w,v)、一个迭代记忆强度参数M，每个并行的深度柔性发射网络训练步骤具体包括如下步骤：Training steps: The purpose is to establish a deep parallel flexible transmission network model, establish a voltage and current Fourier transform matrix model and a data set corresponding to the name of the power terminal to train the model in parallel; when training the fingerprint data of the power terminal, use the flexible transmission network based on the power terminal. The deep flexible emission network model combined with deep learning uses no less than three layers of neurons in the flexible emission network. And during training, different deep flexible transmission network parameters are used for parallel learning, and the number of parallel models is not less than three. The flexible firing network neuron framework includes two feedforward channel parameters (w, v) and an iterative memory strength parameter M, and each parallel deep flexible firing network training step specifically includes the following steps:

(1)前馈过程：以第t个神经元在第l次迭代为例，前馈通道有两个参数(w,v)，神经元自身的记忆强度有一个迭代参数M，对前馈函数进行复数描述后，将得到以下的前馈过程函数为(1) Feedforward process: Taking the t-th neuron in the l-th iteration as an example, the feedforward channel has two parameters (w, v), and the memory strength of the neuron itself has an iterative parameter M. After the complex number description, the following feedforward process function will be obtained as

Figure BDA0002620070540000021
Figure BDA0002620070540000021

其中

Figure BDA0002620070540000022
为本次迭代输出，
Figure BDA0002620070540000023
为下一次迭代的记忆强度，σ为sigmoid激励函数，有实部函数和虚部函数两部分，α和β均为并行模型的可调参数；in
Figure BDA0002620070540000022
Output for this iteration,
Figure BDA0002620070540000023
is the memory strength of the next iteration, σ is the sigmoid excitation function, which has two parts, the real part function and the imaginary part function, α and β are both adjustable parameters of the parallel model;

(2)反向传播过程：在完成一次迭代后，应根据迭代结果对柔性发射网络参数进行校正。(2) Backpropagation process: After completing one iteration, the parameters of the flexible transmitting network should be corrected according to the iteration results.

首先，对模型梯度进行计算，First, the model gradient is calculated,

Figure BDA0002620070540000024
Figure BDA0002620070540000024

其中

Figure BDA0002620070540000025
为反向传播修正差
Figure BDA0002620070540000026
为点乘操作，
Figure BDA0002620070540000027
为激活的点态导数向量，
Figure BDA0002620070540000028
和
Figure BDA0002620070540000029
为反向传播核心计算方法，包括两条分别与
Figure BDA00026200705400000210
和
Figure BDA00026200705400000211
相关的反向传播管道；in
Figure BDA0002620070540000025
Correct the difference for backpropagation
Figure BDA0002620070540000026
For the point multiplication operation,
Figure BDA0002620070540000027
is the activated point-state derivative vector,
Figure BDA0002620070540000028
and
Figure BDA0002620070540000029
It is the core calculation method of backpropagation, including two
Figure BDA00026200705400000210
and
Figure BDA00026200705400000211
The associated back-propagation pipeline;

以给定步长η对W，V进行修正Correction of W, V with a given step size η

Figure BDA00026200705400000212
Figure BDA00026200705400000212

反向传播过程结束后应计算损失函数E。所述损失函数E为柔性发射网络神经元模型的积分损失函数E(W,V)，其通过对真实值和预测值的差的平方来拟合损失，具体地：The loss function E should be calculated after the backpropagation process is over. The loss function E is the integral loss function E(W, V) of the flexible firing network neuron model, which fits the loss by the square of the difference between the actual value and the predicted value, specifically:

Figure BDA00026200705400000213
Figure BDA00026200705400000213

其中Yt为最终层的输出信号，

Figure BDA00026200705400000214
为真实信号；
Figure BDA00026200705400000215
为由0和1组成的向量，若设备编号为k，则
Figure BDA00026200705400000216
为1，反之为0。where Y t is the output signal of the final layer,
Figure BDA00026200705400000214
is a real signal;
Figure BDA00026200705400000215
is a vector consisting of 0 and 1, if the device number is k, then
Figure BDA00026200705400000216
is 1, otherwise it is 0.

根据损失函数计算结果判断继续迭代或训练已达到预期指标。达到预期指标后对M进行最后一次迭代,According to the calculation result of the loss function, it is judged that the iteration or training has reached the expected index. After reaching the expected index, perform the last iteration on M,

Figure BDA00026200705400000217
Figure BDA00026200705400000217

识别步骤：在完成训练步骤后的深度并行柔性发射网络分类器模型中输入电力终端电压电流波形的傅立叶变换矩阵，不同参数的并行模型得出的并行计算结果按照累加和方式进行判断计算，得到最大可能的电力终端编号，视为识别结果。Recognition step: Input the Fourier transform matrix of the voltage and current waveform of the power terminal into the deep parallel flexible transmission network classifier model after the training step is completed, and the parallel calculation results obtained by the parallel models with different parameters are judged and calculated according to the cumulative sum method, and the maximum value is obtained. Possible power terminal numbers, regarded as identification results.

本发明提出的这种电力终端指纹识别方法，能够对大量的电力终端设备的电压电流波形即电力终端指纹进行学习，并在当这些电力终端设备接入时，准确的将这些设备识别出来。通过引入柔性发射网络模型，相较于麦卡洛克·皮特斯模型，运算速度和预测准确性都有了显著的提高，这在大量电力终端接入的情况下，达到了高精度和低算力需求的要求。通过对电力终端指纹的识别，智能电网能够感知接入电网的电力终端设备信息，从而对智能电网的运行提供更多的参考。The power terminal fingerprint identification method proposed by the present invention can learn the voltage and current waveforms of a large number of power terminal devices, that is, power terminal fingerprints, and accurately identify these power terminal devices when they are connected. By introducing the flexible transmission network model, compared with the McCulloch Peters model, the computing speed and prediction accuracy are significantly improved, which achieves high accuracy and low computing power when a large number of power terminals are connected. demand requirements. Through the identification of power terminal fingerprints, the smart grid can perceive the information of the power terminal equipment connected to the power grid, thereby providing more reference for the operation of the smart grid.

附图说明Description of drawings

图1是本发明方法的总体设计流程图。Fig. 1 is the overall design flow chart of the method of the present invention.

图2是本发明方法的柔性发射网络示意图。FIG. 2 is a schematic diagram of a flexible transmission network of the method of the present invention.

具体实施方式Detailed ways

本发明提出的一种深度并行柔性发射网络的电力终端指纹识别方法，结合附图详细说明如下：A power terminal fingerprint identification method of a deep parallel flexible transmission network proposed by the present invention is described in detail as follows with reference to the accompanying drawings:

图1是本发明方法的总体设计流程图。首先，从大量的电力终端设备中获取不同运行条件下的电压、电流波形，通过深度并行柔性发射网络对波形特征进行学习，实现介入新设备产生新波形后的电力终端指纹识别，具体通过以下步骤实现。Fig. 1 is the overall design flow chart of the method of the present invention. First, the voltage and current waveforms under different operating conditions are obtained from a large number of power terminal equipment, and the waveform characteristics are learned through the deep parallel flexible transmission network, so as to realize the fingerprint identification of the power terminal after the new equipment is involved to generate a new waveform. The specific steps are as follows. accomplish.

步骤S1，对需要进行学习的电力终端设备进行工况采集，将这些电力终端设备连接至电压和电流监测设备上，对其施加不同的工况，得到电力终端设备在不同工作条件下的电压和电流波形。对波形进行高频采集，并存储形成测试数据。Step S1, collect the working conditions of the power terminal equipment that needs to be learned, connect these power terminal equipment to the voltage and current monitoring equipment, apply different working conditions to it, and obtain the voltage and voltage of the power terminal equipment under different working conditions. current waveform. The waveform is collected at high frequency and stored to form the test data.

步骤S2，对步骤S1中高频采集到的电压电流数据进行处理，对其进行离散傅里叶分解运算，得到电压电流波形的各次谐波幅值和相位信息，对其整合形成一个4n的矩阵，即视为电力终端设备的指纹。由于在电网中负荷往往呈现近正弦波形特性，高次谐波含量逐渐降低，因此电力终端指纹矩阵可以合理弃值，忽略较高频率的分解值，提高后续步骤的运算速度。Step S2, process the voltage and current data collected by the high frequency in step S1, perform discrete Fourier decomposition operation on it, obtain the amplitude and phase information of each harmonic of the voltage and current waveform, and integrate them to form a 4n matrix. , which is regarded as the fingerprint of the power terminal equipment. Since the load in the power grid often presents a near-sinusoidal waveform characteristic, and the high-order harmonic content gradually decreases, the power terminal fingerprint matrix can reasonably discard the value, ignore the decomposition value of higher frequency, and improve the operation speed of the subsequent steps.

步骤S3，使用步骤S2处理的电力终端指纹数据对深度并行柔性发射网络进行训练，柔性发射网络是在麦卡洛克·皮特斯模型的基础上改进而来，它引入了新的仿生学原理，根据生物学对神经元的研究结果进行改进，将前馈通道的参数改进为2个，并引入了第3个参数来模拟神经元的记忆强度，改进后的模型有了更快的目标拟合速度和更高的拟合精度。在执行训练之前，首先应合理设置深度柔性发射网络的深度和迭代参数，深度学习的神经元个数应在3个以上，并使用并行计算对不同参数或深度的深度柔性发射网络模型进行训练。Step S3, use the power terminal fingerprint data processed in step S2 to train the deep parallel flexible transmission network. The flexible transmission network is improved on the basis of the McCulloch Peters model. It introduces a new bionics principle, according to Biology improves the research results of neurons, improves the parameters of the feedforward channel to 2, and introduces a third parameter to simulate the memory strength of neurons, the improved model has a faster target fitting speed and higher fitting accuracy. Before performing training, the depth and iteration parameters of the deep flexible emission network should be reasonably set, and the number of neurons in deep learning should be more than 3, and parallel computing should be used to train deep flexible emission network models with different parameters or depths.

柔性发射网络神经元框架包括两个前馈通道参数(w,v)、一个迭代记忆强度参数M，每个并行的深度柔性发射网络训练步骤具体包括如下步骤：The neuron framework of the flexible firing network includes two feedforward channel parameters (w, v) and an iterative memory strength parameter M. Each parallel deep flexible firing network training step specifically includes the following steps:

(1)前馈过程：以第t个神经元为例，前馈通道有两个参数(w,v)，神经元自身的记忆强度有一个迭代参数M，对前馈函数进行复数描述后，将得到以下的前馈过程函数为(1) Feedforward process: Taking the t-th neuron as an example, the feedforward channel has two parameters (w, v), and the memory strength of the neuron itself has an iterative parameter M. After the complex number description of the feedforward function, The following feedforward process function will be obtained as

Figure BDA0002620070540000041
Figure BDA0002620070540000041

其中

Figure BDA0002620070540000042
为本次迭代输出，
Figure BDA0002620070540000043
为下一次迭代的记忆强度，σ为sigmoid激励函数，有实部函数和虚部函数两部分，α和β均为并行模型的可调参数；in
Figure BDA0002620070540000042
Output for this iteration,
Figure BDA0002620070540000043
is the memory strength of the next iteration, σ is the sigmoid excitation function, which has two parts, the real part function and the imaginary part function, α and β are both adjustable parameters of the parallel model;

(2)反向传播过程：在完成一次迭代后，应根据迭代结果对柔性发射网络参数进行校正，首先对模型梯度进行计算，(2) Backpropagation process: After completing one iteration, the parameters of the flexible transmitting network should be corrected according to the iteration results, and the model gradient should be calculated first.

Figure BDA0002620070540000044
Figure BDA0002620070540000044

其中

Figure BDA0002620070540000045
为反向传播修正差，
Figure BDA0002620070540000046
为点乘操作，
Figure BDA0002620070540000047
为激活的点态导数向量
Figure BDA0002620070540000048
Figure BDA00026200705400000415
和
Figure BDA0002620070540000049
为反向传播核心计算方法。包括两条分别与
Figure BDA00026200705400000410
和
Figure BDA00026200705400000411
相关的反向传播管道；in
Figure BDA0002620070540000045
Correct the difference for backpropagation,
Figure BDA0002620070540000046
For the point multiplication operation,
Figure BDA0002620070540000047
is the activated point-state derivative vector
Figure BDA0002620070540000048
Figure BDA00026200705400000415
and
Figure BDA0002620070540000049
The core calculation method for backpropagation. including two
Figure BDA00026200705400000410
and
Figure BDA00026200705400000411
The associated back-propagation pipeline;

其中：in:

Figure BDA00026200705400000412
Figure BDA00026200705400000412

Figure BDA00026200705400000413
Figure BDA00026200705400000413

其中：in:

Figure BDA00026200705400000414
Figure BDA00026200705400000414

其中：in:

Figure BDA0002620070540000051
Figure BDA0002620070540000051

计算结束后，以给定步长η对W，V进行修正After the calculation, correct W and V with a given step size η

Figure BDA0002620070540000052
Figure BDA0002620070540000052

反向传播过程结束后应计算损失函数E。所述损失函数E为柔性发射网络神经元模型的积分损失函数E(W,V)，其通过对真实值和预测值的差平方来拟合损失，具体地：The loss function E should be calculated after the backpropagation process is over. The loss function E is the integral loss function E(W, V) of the flexible firing network neuron model, which fits the loss by the square of the difference between the actual value and the predicted value, specifically:

Figure BDA0002620070540000053
Figure BDA0002620070540000053

其中Yt为最终层的输出信号，

Figure BDA0002620070540000054
为真实信号；
Figure BDA0002620070540000055
为由0和1组成的向量，若设备编号为k，则
Figure BDA0002620070540000056
为1，反之为0。where Y t is the output signal of the final layer,
Figure BDA0002620070540000054
is a real signal;
Figure BDA0002620070540000055
is a vector consisting of 0 and 1, if the device number is k, then
Figure BDA0002620070540000056
is 1, otherwise it is 0.

根据损失函数计算结果判断继续迭代或训练已达到预期指标。达到预期指标后对M进行最后一次迭代,According to the calculation result of the loss function, it is judged that the iteration or training has reached the expected index. After reaching the expected index, perform the last iteration on M,

Figure BDA0002620070540000057
Figure BDA0002620070540000057

步骤S4，在各个并行模型损失函数达到要求之后，即可对电力终端指纹进行识别。将电力终端设备的电压电流波形进行采集，对其进行离散傅里叶变换，将变换结果输入到完成学习的深度并行柔性发射网络模型中，在各个并行模型中进行计算，将各个并行结果进行相加，得到可能最大的结果，即视为识别结果。Step S4, after each parallel model loss function meets the requirements, the fingerprint of the power terminal can be identified. Collect the voltage and current waveforms of the power terminal equipment, perform discrete Fourier transformation on them, input the transformation results into the deep parallel flexible transmission network model that has completed the learning, perform calculations in each parallel model, and compare each parallel result. Add to get the largest possible result, which is regarded as the recognition result.

图2是本发明方法的柔性发射网络示意图。本发明提出的电力终端指纹识别方法通过引入柔性发射网络模型，相较于麦卡洛克·皮特斯模型，运算速度和预测准确性都有了显著的提高，这在实际工程中即大量电力终端接入的情况下，达到了高精度和低算力需求的要求。通过对电力终端指纹的识别，智能电网能够感知接入电网的电力终端设备信息，从而对智能电网的运行提供更多的参考。FIG. 2 is a schematic diagram of a flexible transmission network of the method of the present invention. Compared with the McCulloch-Peters model, the power terminal fingerprint identification method proposed by the present invention has significantly improved operation speed and prediction accuracy by introducing a flexible transmission network model. In the case of input, the requirements of high precision and low computing power are met. Through the identification of power terminal fingerprints, the smart grid can perceive the information of the power terminal equipment connected to the power grid, thereby providing more reference for the operation of the smart grid.

Claims (7)

1.一种深度并行柔性发射网络的电力终端指纹识别方法，其特征在于，能用一种深度并行柔性发射网络对电力终端指纹进行学习，在电力终端接入电力系统时对其进行识别；该方法在使用过程中的主要步骤为：1. a power terminal fingerprint identification method of a deep parallel flexible transmission network, it is characterized in that, can use a kind of deep parallel flexible transmission network to learn the power terminal fingerprint, when the power terminal is connected to the power system, it is identified; the The main steps in the process of using the method are: (1)用电数据获取步骤：采集各个电力终端在单独运行时处于不同工作状态下的电压和电流的用电数据；(1) Power consumption data acquisition step: collect the power consumption data of the voltage and current of each power terminal in different working states when running alone; (2)数据预处理步骤：对步骤(1)中得到的电压和电流波形进行离散傅立叶变换形成4n矩阵作为电力终端指纹特征；(2) data preprocessing step: carry out discrete Fourier transform to the voltage and current waveforms obtained in step (1) to form a 4n matrix as a power terminal fingerprint feature; (3)训练步骤：建立深度并行柔性发射网络模型，建立电压、电流傅立叶变换矩阵模型与电力终端名称对应的数据集对该模型进行并行训练；(3) Training step: establish a deep parallel flexible transmission network model, establish a voltage and current Fourier transform matrix model and a data set corresponding to the name of the power terminal to perform parallel training on the model; (4)识别步骤：在完成训练步骤后的深度并行柔性发射网络分类器模型中输入电力终端电压电流波形的离散傅立叶变换矩阵，经并行系统判断后，输出电力终端指纹匹配结果。(4) Identification step: input the discrete Fourier transform matrix of the voltage and current waveforms of the power terminal in the deep parallel flexible transmission network classifier model after the training step is completed, and output the fingerprint matching result of the power terminal after being judged by the parallel system. 2.如权利要求1所述的深度并行柔性发射网络的电力终端指纹识别方法，其特征在于，所述步骤(2)中所述电力终端指纹特征为电力终端不同运行状态下的电压和电流波形，对其进行离散傅立叶变换得到的4n矩阵即为该特征的数学模型。2. the power terminal fingerprint identification method of the deep parallel flexible transmission network as claimed in claim 1 is characterized in that, described in the step (2), the power terminal fingerprint feature is the voltage and current waveforms under different operating states of the power terminal , and the 4n matrix obtained by discrete Fourier transform is the mathematical model of the feature. 3.如权利要求1所述的深度并行柔性发射网络的电力终端指纹识别方法，其特征在于，所述步骤(3)中对电力终端指纹数据进行训练时，使用基于柔性发射网络和深度学习结合的深度柔性发射网络模型；且在训练时，使用不同的深度柔性发射网络参数进行并行学习。3. the power terminal fingerprint identification method of the deep parallel flexible transmission network as claimed in claim 1 is characterized in that, when the power terminal fingerprint data is trained in the described step (3), use the combination based on the flexible transmission network and deep learning The deep flexible transmission network model of , and during training, different deep flexible transmission network parameters are used for parallel learning. 4.如权利要求1所述的深度并行柔性发射网络的电力终端指纹识别方法，其特征在于，所述步骤(4)中对电力终端指纹数据进行识别时，使用权利要求3所述的深度柔性发射网络进行识别，并通过并行模型输出多个预测值，按照累加和方式进行判断并确定识别结果。4. the power terminal fingerprint identification method of the deep parallel flexible transmission network as claimed in claim 1 is characterized in that, when the power terminal fingerprint data is identified in the described step (4), the deep flexible according to claim 3 is used. The transmitting network performs identification, and outputs multiple predicted values through the parallel model, and judges and determines the identification result according to the cumulative sum method. 5.如权利要求1所述的深度并行柔性发射网络的电力终端指纹识别方法，其深度特征在于，使用的柔性发射网络神经元层数不小于三层，其并行特征在于，并行模型数不少于三个，所述柔性发射网络神经元框架包括两个前馈通道参数(w，v)、一个迭代记忆强度参数M和损失函数E；所述损失函数E为柔性发射网络神经元模型的积分损失函数E(W，V)，其通过对真实值和预测值的差的平方来拟合损失，具体地：5. The power terminal fingerprint identification method of the deep parallel flexible transmission network as claimed in claim 1, its depth feature is that the number of neuron layers of the flexible transmission network used is not less than three layers, and its parallel feature is that the number of parallel models is quite large For three, the flexible firing network neuron framework includes two feedforward channel parameters (w, v), an iterative memory strength parameter M and a loss function E; the loss function E is the integral of the flexible firing network neuron model A loss function E(W, V) that fits the loss by squaring the difference between the true and predicted values, specifically:
Figure FDA0002620070530000011
Figure FDA0002620070530000011
 其中Yt为最终层的输出信号；
Figure FDA0002620070530000012
为真实信号；
Figure FDA0002620070530000013
为由0和1组成的向量，若设备编号为k，则
Figure FDA0002620070530000014
为1，反之为0。where Y t is the output signal of the final layer;
Figure FDA0002620070530000012
is a real signal;
Figure FDA0002620070530000013
is a vector consisting of 0 and 1, if the device number is k, then
Figure FDA0002620070530000014
is 1, otherwise it is 0. 6.如权利要求1所述的深度并行柔性发射网络的电力终端指纹识别方法，其特征在于，每个并行的深度柔性发射网络训练步骤具体包括如下步骤，6. The power terminal fingerprint identification method of the deep parallel flexible transmission network as claimed in claim 1, is characterized in that, each parallel deep flexible transmission network training step specifically comprises the following steps, (1)前馈过程：以第t个神经元在第l次迭代为例，前馈通道有两个参数(w，v)，神经元自身的记忆强度有一个迭代参数M，对前馈函数进行复数描述后，将得到以下的前馈过程函数为(1) Feedforward process: Taking the t-th neuron in the l-th iteration as an example, the feed-forward channel has two parameters (w, v), and the memory strength of the neuron itself has an iterative parameter M. After the complex number description, the following feedforward process function will be obtained as
Figure FDA0002620070530000021
Figure FDA0002620070530000021
 其中
Figure FDA0002620070530000022
为本次迭代输出，
Figure FDA0002620070530000023
为下一次迭代的记忆强度，σ为sigmoid激励函数，有实部函数和虚部函数两部分，α和β均为并行模型的可调参数；in
Figure FDA0002620070530000022
Output for this iteration,
Figure FDA0002620070530000023
is the memory strength of the next iteration, σ is the sigmoid excitation function, which has two parts, the real part function and the imaginary part function, α and β are both adjustable parameters of the parallel model; (2)反向传播过程：在完成一次迭代后，应根据迭代结果对柔性发射网络参数进行校正，首先对模型梯度进行计算，(2) Backpropagation process: After completing one iteration, the parameters of the flexible transmitting network should be corrected according to the iteration results, and the model gradient should be calculated first.
Figure FDA0002620070530000024
Figure FDA0002620070530000024
 其中
Figure FDA0002620070530000025
为反向传播修正差，
Figure FDA0002620070530000026
为点乘操作，
Figure FDA0002620070530000027
为激活的点态导数向量，
Figure FDA0002620070530000028
和
Figure FDA0002620070530000029
为反向传播核心计算方法，包括两条分别与
Figure FDA00026200705300000210
和
Figure FDA00026200705300000211
相关的反向传播管道；in
Figure FDA0002620070530000025
Correct the difference for backpropagation,
Figure FDA0002620070530000026
For the point multiplication operation,
Figure FDA0002620070530000027
is the activated point-state derivative vector,
Figure FDA0002620070530000028
and
Figure FDA0002620070530000029
It is the core calculation method of backpropagation, including two
Figure FDA00026200705300000210
and
Figure FDA00026200705300000211
The associated back-propagation pipeline; 以给定步长η对W，V进行修正，Correcting W, V with a given step size η,
Figure FDA00026200705300000212
Figure FDA00026200705300000212
 反向传播过程结束，根据损失函数计算结果判断继续迭代或训练已达到预期指标，达到预期后对M进行最后一次迭代，At the end of the backpropagation process, according to the calculation result of the loss function, it is judged that the iteration continues or the training has reached the expected index. After reaching the expectation, the last iteration is performed on M,
Figure FDA00026200705300000213
Figure FDA00026200705300000213
 7.如权利要求1所述的深度并行柔性发射网络的电力终端指纹识别方法，其特征在于，将电力终端运行的实时电压和电流波形进行离散傅立叶变换形成4n矩阵后，送入如权利要求6所述的已经训练完成的深度并行柔性发射网络中进行计算，得出的并行计算结果按照累加和方式进行判断计算，得到最大可能的电力终端编号，视为识别结果。7. the power terminal fingerprint identification method of deep parallel flexible transmission network as claimed in claim 1, is characterized in that, after the real-time voltage and current waveform of power terminal operation are carried out discrete Fourier transform to form 4n matrix, send into as claimed in claim 6 The calculation is performed in the deep parallel flexible transmission network that has been trained, and the obtained parallel calculation results are judged and calculated according to the cumulative sum method, and the maximum possible power terminal number is obtained, which is regarded as the identification result.
CN202010780676.2A 2020-08-06 2020-08-06 Electric power terminal fingerprint identification method of deep parallel flexible transmission network Active CN112016684B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010780676.2A CN112016684B (en) 2020-08-06 2020-08-06 Electric power terminal fingerprint identification method of deep parallel flexible transmission network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010780676.2A CN112016684B (en) 2020-08-06 2020-08-06 Electric power terminal fingerprint identification method of deep parallel flexible transmission network

Publications (2)

Publication Number Publication Date
CN112016684A true CN112016684A (en) 2020-12-01
CN112016684B CN112016684B (en) 2022-09-02

Family

ID=73499055

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010780676.2A Active CN112016684B (en) 2020-08-06 2020-08-06 Electric power terminal fingerprint identification method of deep parallel flexible transmission network

Country Status (1)

Country Link
CN (1) CN112016684B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113205177A (en) * 2021-04-25 2021-08-03 广西大学 Electric power terminal identification method based on incremental collaborative attention mobile convolution

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101900761A (en) * 2009-11-05 2010-12-01 中国航天科技集团公司第五研究院第五一四研究所 High-accuracy non-integer-period sampled harmonic analysis and measurement method
CN102129013A (en) * 2011-01-21 2011-07-20 昆明理工大学 Distribution network fault location method utilizing natural frequency and artificial neural network
CN107045649A (en) * 2014-07-21 2017-08-15 徐志强 Simulative neural network with short-term memory and long-term memory
CN109727678A (en) * 2017-10-28 2019-05-07 徐志强 Variable neuron, cynapse simulator and simulated substrate neural circuitry
CN110503188A (en) * 2019-08-01 2019-11-26 三星（中国）半导体有限公司 Method and device for implementing deep neural network model based on heterogeneous model graph
CN110705456A (en) * 2019-09-29 2020-01-17 华南理工大学 Micro motor abnormity detection method based on transfer learning
CN110738253A (en) * 2019-10-14 2020-01-31 南京信息工程大学 A short-term wind power prediction method based on FCM and AFSA-Elman

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101900761A (en) * 2009-11-05 2010-12-01 中国航天科技集团公司第五研究院第五一四研究所 High-accuracy non-integer-period sampled harmonic analysis and measurement method
CN102129013A (en) * 2011-01-21 2011-07-20 昆明理工大学 Distribution network fault location method utilizing natural frequency and artificial neural network
CN107045649A (en) * 2014-07-21 2017-08-15 徐志强 Simulative neural network with short-term memory and long-term memory
CN109727678A (en) * 2017-10-28 2019-05-07 徐志强 Variable neuron, cynapse simulator and simulated substrate neural circuitry
CN110503188A (en) * 2019-08-01 2019-11-26 三星（中国）半导体有限公司 Method and device for implementing deep neural network model based on heterogeneous model graph
CN110705456A (en) * 2019-09-29 2020-01-17 华南理工大学 Micro motor abnormity detection method based on transfer learning
CN110738253A (en) * 2019-10-14 2020-01-31 南京信息工程大学 A short-term wind power prediction method based on FCM and AFSA-Elman

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
JEAN-BAPTISTEDORÉ等: "Waveform contenders for 5G: Description, analysis and comparison", 《PHYSICAL COMMUNICATION》 *
MA XIAOCHEN等: "Deep Flexible Transmitter Networks for Non-Intrusive Load Monitoring of Power Distribution Networks", 《IEEE ACCESS》 *
SHAO-QUN ZHANG等: "Flexible Transmitter Network", 《ARXIV:2004.03839V2 [CS.NE]》 *
庞昊等: "基于多神经网络融合的短期负荷预测方法", 《电力自动化设备》 *
黄丹: "基于小波和神经网络算法的电力系统短期负荷预测的研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113205177A (en) * 2021-04-25 2021-08-03 广西大学 Electric power terminal identification method based on incremental collaborative attention mobile convolution
CN113205177B (en) * 2021-04-25 2022-03-25 广西大学 Electric power terminal identification method based on incremental collaborative attention mobile convolution

Also Published As

Publication number Publication date
CN112016684B (en) 2022-09-02

Similar Documents

Publication Publication Date Title
CN111860982B (en) VMD-FCM-GRU-based wind power plant short-term wind power prediction method
CN109597043B (en) Radar signal recognition method based on quantum particle swarm convolutional neural network
CN109884419B (en) An online fault diagnosis method for power quality of smart grid
CN109145516B (en) A fault identification method for analog circuits based on an improved extreme learning machine
CN103018673A (en) Method for predicating life of aerospace Ni-Cd storage battery based on improved dynamic wavelet neural network
CN110244205A (en) A Partial Discharge Recognition Method of Cable Accessories Based on Convolutional Neural Network under Small Data Amount
CN110443724B (en) Electric power system rapid state estimation method based on deep learning
CN101587546A (en) Identification method of lower limb action pattern based on supporting vector multivariate classification
CN113553755B (en) Power system state estimation method, device and equipment
CN106053988A (en) Inverter fault diagnosis system and method based on intelligent analysis
CN113804997B (en) Voltage sag source positioning method based on bidirectional WaveNet deep learning
CN115374710A (en) Wind speed prediction method based on improved seagull optimization algorithm and multi-core extreme learning machine
CN116224081A (en) Battery SOH estimation method based on LOF-Pearson detection multidimensional feature vector
CN108985313A (en) AGC system discrimination method based on big data and Bayesian neural network
CN112016684B (en) Electric power terminal fingerprint identification method of deep parallel flexible transmission network
CN112014757A (en) Battery SOH estimation method integrating capacity increment analysis and genetic wavelet neural network
CN112836876A (en) A deep learning-based distribution network line load forecasting method
CN109459609B (en) Distributed power supply frequency detection method based on artificial neural network
CN108805206A (en) Improved L SSVM establishing method for analog circuit fault classification
CN118585911B (en) Active power distribution network short-circuit fault identification method, device and storage medium
CN115859702A (en) A method and system for demagnetization fault diagnosis of permanent magnet synchronous wind turbine based on convolutional neural network
CN114841266A (en) A voltage sag identification method based on triple prototype network with small samples
CN113158134B (en) Method, device and storage medium for constructing non-invasive load identification model
CN113205177A (en) Electric power terminal identification method based on incremental collaborative attention mobile convolution
CN117353286A (en) A short-term load forecasting method based on multiple similar days and stacking learning

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
OL01 Intention to license declared
OL01 Intention to license declared
EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20201201

Assignee: Guangxi Nanning Xinyi Painting Technology Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980034788

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241212

EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20201201

Assignee: GUANGXI SENGE AUTOMATION TECHNOLOGY Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980037759

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241215

Application publication date: 20201201

Assignee: Guangxi Sustainable E-Commerce Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980037727

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241215

Application publication date: 20201201

Assignee: Guangxi aid Energy Technology Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980037700

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241215

Application publication date: 20201201

Assignee: GUANGXI ZHONGLIAN ZHIHAO TECHNOLOGY Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980033392

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241213

EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20201201

Assignee: Nanning Yanxiang Industrial Supply Chain Management Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980038691

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241216

Application publication date: 20201201

Assignee: Guangxi Guiren Energy Saving Technology Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980038688

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241216

Application publication date: 20201201

Assignee: GUANGXI CHAOXING SOLAR ENERGY TECHNOLOGY Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980038685

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241216

Application publication date: 20201201

Assignee: Nanning Yanxiang Equipment Technology Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980037772

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241215

Application publication date: 20201201

Assignee: Guangxi Gaonong Intelligent Technology Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980037739

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241215

EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20201201

Assignee: Nanning Yanxiang Intelligent Technology Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980040428

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Common License

Record date: 20241220

EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20201201

Assignee: Guangxi Zegang Technology Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2024980044615

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Open License

Record date: 20241231

EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20201201

Assignee: Nanning Zhongping electronic weighing apparatus Co.,Ltd.

Assignor: GUANGXI University

Contract record no.: X2025980000158

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Open License

Record date: 20250106

Application publication date: 20201201

Assignee: GUANGXI LANGJIE WISDOM TECHNOLOGY DEVELOPMENT CO.,LTD.

Assignor: GUANGXI University

Contract record no.: X2025980000126

Denomination of invention: A deep parallel flexible transmission network for power terminal fingerprint recognition method

Granted publication date: 20220902

License type: Open License

Record date: 20250106