CN102915511A - Safety monitoring method for neural network model of power-loaded chaotic phase space - Google Patents

Safety monitoring method for neural network model of power-loaded chaotic phase space Download PDF

Info

Publication number
CN102915511A
CN102915511A CN2012103607943A CN201210360794A CN102915511A CN 102915511 A CN102915511 A CN 102915511A CN 2012103607943 A CN2012103607943 A CN 2012103607943A CN 201210360794 A CN201210360794 A CN 201210360794A CN 102915511 A CN102915511 A CN 102915511A
Authority
CN
China
Prior art keywords
phase space
neural network
load
output
value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN2012103607943A
Other languages
Chinese (zh)
Inventor
李眉眉
第宝锋
黄正文
柯玲
丁晶
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.)
Sichuan University
Original Assignee
Sichuan 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 Sichuan University filed Critical Sichuan University
Priority to CN2012103607943A priority Critical patent/CN102915511A/en
Publication of CN102915511A publication Critical patent/CN102915511A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention relates to a safety monitoring method for a neural network model of a power-loaded chaotic phase space. The safety monitoring method is a physical modeling analysis method which takes a power daily load as a research object, reconstructs a phase space based on a chaotic theory, discusses a nonlinear prediction scheme in a load phase space and then apply the prediction scheme to practical prediction and safety monitoring. The safety monitoring method comprises the following steps: A, acquiring daily load data of a power system to form a time sequence; B, calculating self-correlation functions of the sequence; C, calculating saturated correlation dimensions of the daily load sequence to prove a chaotic characteristic of the sequence; D, establishing a multi-dimensional phase space of the power load to form a studying sample and a teacher value; E, determining a neural network structure; F, processing normalization of the data; G, studying the network; H, outputting predicted values; and I, restoring the neural network output values to practical load values. With adoption of the safety monitoring method, the power load is proved to have the chaotic characteristic; prediction accuracy of the model is high; and the change rule of the power daily load can be effectively monitored so as to guarantee safe and economical running of the power system.

Description

A kind of chaos phase space neural network model safety monitoring method of electric load
Technical field
The present invention relates to a kind of chaos phase space neural network model safety monitoring method of electric load, namely take the electric power daily load as research object, based on chaology, reconstruct electric load phase space, and in conjunction with the model and algorithm of neural network, how research extracts the dynamic information feature of electric load, inquires into the nonlinear prediction scheme in the load phase space, then be applied to the physical modeling analytical approach in actual prediction and the safety monitoring, belong to power domain.
Background technology
Power system load data as water resources development, distribute rationally, the important evidence of reservoir operation.Load forecast plays a very important role the safety and economic operation of electric system.The daily load safe prediction of electric system is that electrical network is to arrange to purchase electricity plan and the service of transmission of electricity the establishment of the project.The electric load time series that obtains in the practice presents complicacy, uncertainty, nonlinear characteristics.And neural network can be shone upon the nonlinear relationship of any complexity, on the connection weight that the implicit feature of sample and rule is distributed in neural network by study.Based on chaology, reconstruct electric load phase space, the coupled neural network model, how research extracts the dynamic information feature of electric load, inquire into the nonlinear prediction scheme in the load phase space, it is very significant being applied in actual prediction and the safety monitoring again.
Summary of the invention
The chaos phase space neural network model safety monitoring method of a kind of electric load that the present invention proposes, its technical scheme is as follows.
1, gathers electric system daily load data and consist of the daily load time series.
2, ask for the autocorrelation function of Sichuan Province's electric power daily load.
3, calculate the saturated correlation dimension of daily load sequence, thus the chaotic characteristic of proof sequence.
4, set up the multidimensional phase space of electric load chaos time sequence, consist of learning sample and teacher value.
Determine optimum delay time with the autocorrelation function method τ, determine the best dimension that embeds with saturated correlation dimension method m, the phase space of reconfiguration system:
Figure 2012103607943100002DEST_PATH_IMAGE001
Wherein N=n-( M-1) * τLength for sequence vector.
5, determine neural network structure.Input layer dimension by network is m, i.e. the embedding dimension of chaos phase space, output dimension are one dimension (i.e. the output of prediction), the Hidden unit dimension is determined in network training study by trial and error.
6, the normalized of data.All sample values, teacher's value and output valve with unified method for normalizing, are used conversion
Figure 2012103607943100002DEST_PATH_IMAGE002
Load is scaled [0,1] interval value.
7, the study of network.Any two phase points in the chaos phase space
Figure 2012103607943100002DEST_PATH_IMAGE003
With
Figure 2012103607943100002DEST_PATH_IMAGE004
(wherein TBe leading time), their nonlinear Evolution funtcional relationships in phase space
Figure 2012103607943100002DEST_PATH_IMAGE005
, come match by artificial neural network.Although prediction output point
Figure 971084DEST_PATH_IMAGE004
For mN dimensional vector n, but in fact ( m-1) dimension is known quantity, gets for the sake of simplicity network and is output as
Figure 22216DEST_PATH_IMAGE004
mIndividual component.(see figure 1)
To the neuron j of any one non-linear unit, its input/output relation has:
Figure 2012103607943100002DEST_PATH_IMAGE006
（1）
In the formula, xBe neuronic input, ω is connection weight, and θ is threshold value, yBe this neuronic output. fBe excitation function, commonly used is the Sigmoid function, namely
Figure 2012103607943100002DEST_PATH_IMAGE007
（2）
Error backpropagation algorithm (BP algorithm) is pressed in the connection weight of neural network and the study of threshold value, and this is a kind of teacher's of having learning algorithm.Learning process comprises positive and negative two processes, at first by the output of each layer of formula (1) forward computational grid, calculate again the error between output and the teacher value, utilize error oppositely to revise connection weight and threshold value, repeat above two processes until the error of output and sample teacher value is little of a certain setting value ε.
Suppose that the input learning sample is kIt is individual,
Figure 2012103607943100002DEST_PATH_IMAGE008
, the predicted value of output is
Figure 2012103607943100002DEST_PATH_IMAGE009
, corresponding teacher's value is
Figure 2012103607943100002DEST_PATH_IMAGE010
, objective function is:
Figure 2012103607943100002DEST_PATH_IMAGE011
（3）
Gradient descent method is adopted in the correction of connection weight and threshold value, and in order to prevent concussion and accelerating convergence, this paper adopts the algorithm of additional momentum item:
Figure 2012103607943100002DEST_PATH_IMAGE012
（4）
In the formula, Connection weight or the threshold value of difference input layer, hidden layer and output layer,
Figure 2012103607943100002DEST_PATH_IMAGE014
Be iterations, η is training speed,
Figure 2012103607943100002DEST_PATH_IMAGE015
Be the modification value,
Figure 2012103607943100002DEST_PATH_IMAGE016
Be factor of momentum.
8, forecast model.The newspaper phase point that rises of chaos phase space is input to neural network, and the output of network is predicted value.
9, use conversion
Figure 2012103607943100002DEST_PATH_IMAGE017
The value interval to neural network output [0,1] is reduced into the actual load value.
Beneficial effect of the present invention:
ⅰ. the present invention illustrates that the electric power daily load has chaotic characteristic.By calculating the saturated correlation dimension of electric power daily load sequence, thereby proved the chaotic characteristic of daily load sequence.Reconstruct daily load phase space consists of neural network model, utilizes the BP algorithm to carry out the study correction of network, and in phase space the daily load sequence is carried out nonlinear prediction.
ⅱ. the chaos phase space neural network model safety monitoring method of a kind of electric load that the present invention proposes, utilize the chaos reconstruction Phase Space Theory fully to excavate the intrinsic characteristic that the electric power daily load changes, and the non-linear rule of coupled neural network mapping phase point evolution, precision of prediction is satisfactory, can effectively monitor electric power daily load Changing Pattern, thereby provide important evidence for the safety and economic operation of electric system.
ⅲ. in order to strengthen the Generalization Ability of model, further improve precision of prediction, can in model, consider weather conditions, festivals or holidays influence factor etc., make model higher at the precision of prediction of the singular point of Power system load data.
(4) description of drawings:
The neural network structure figure of Fig. 1 chaos phase space
Fig. 2 Sichuan Province electric power daily load time series
The autocorrelation function graph of Fig. 3 Sichuan Province electric power daily load
The ln of Fig. 4 Sichuan Province electric power daily load C( r)-ln rGraph of a relation
The saturated correlation dimension figure of Fig. 5 Sichuan Province electric power daily load
(5) embodiment:
1, chooses Sichuan Province the whole province electric system daily load time series (Dec in August, 1998 to calendar year 2001), unit: MW; (see figure 2).
2, ask for the autocorrelation function of Sichuan Province's electric power daily load.Autocorrelation function is for the first time less than 0.368(1/e) time corresponding when stagnant, choose time delay of electric power daily load τ=60 days; (see figure 3).
3, judge the chaotic characteristic of Sichuan Province's electric system daily load sequence.Calculate its correlation dimension (see figure 4), saturated correlation dimension with the G-P algorithm from the daily load time series m=7(sees Fig. 5).
4, reconstruct electric system daily load seasonal effect in time series phase space.The embedding dimension of selecting correlation dimension to occur when saturated is the embedding dimension of phase space reconstruction m=7, time delay τ=60, such choosing makes the embedding window width
Figure 2012103607943100002DEST_PATH_IMAGE018
Close to the average period (≈ 360 days) of daily load.
5, determine neural network structure.By the input layer number that embeds dimension and determine network, the hidden neuron number is preferred in the study of network with trial and error, and model structure is 7-7-1, and network is output as the electric load of prediction day, and be 1 day the forecast period of prediction.
6, Sichuan Province's electric power daily load data will be carried out normalized.All sample values, teacher's value and output valve with unified method for normalizing, are used conversion
Figure 671415DEST_PATH_IMAGE002
Daily load is scaled [0,1] interval value.
7, from Sichuan Province's electric power daily load data, to choose the load in year November in August, 1998 to 1999 and do the sample data neural network training, the electric power daily load in November, 1999 is as model testing, forecast period T=1 heaven-made short-term forecasting.Reconstruct load 7 dimension phase spaces then have 97 phase points and do training sample.Select training speed η=0.03 in the BP network training, factor of momentum =0.8, train 1000 times training permissible error ε=0.01.
8, the study of network.The initial value of network connection power is set at random, by formula (1) to the output of formula (4) computational grid and revise weights, until error is controlled at allowed band or iterations reaches predetermined value.
9, will play the newspaper phase point and be input to neural network, the output of network is the predicted value of Sichuan Province's electric power daily load.
10, actual output daily load data conversion
Figure 629324DEST_PATH_IMAGE017
Reduce.
11, the qualification rate of chaos phase space Neural Network model predictive (relative error less than 10% number percent) is 97.938%, and the average relative error of model is 3.048%, and maximum relative error is 10.741%.The prediction qualification rate of model (relative error less than 10% number percent) is 100%, and average relative error is 2.886%, and maximum relative error is 7.813%.The fitting precision of model and precision of prediction are all comparatively desirable.The result of model prediction is as shown in table 1.
Table 1 chaos phase space Neural Network model predictive result

Claims (6)

1. the chaos phase space neural network model safety monitoring method of an electric load, its described method characteristic is: sequentially comprise the steps or feature:
A. gather electric system daily load data and consist of the daily load time series;
B. ask for the autocorrelation function of Sichuan Province's electric power daily load;
C. calculate the saturated correlation dimension of daily load sequence, thus the chaotic characteristic of proof sequence;
D. set up the multidimensional phase space of electric load chaos time sequence, consist of learning sample and teacher value;
E. determine neural network structure;
F. the normalized of data;
G. the study of network;
H. forecast model;
I. use conversion The value interval to neural network output [0,1] is reduced into the actual load value.
2. the chaos phase space neural network model safety monitoring method of the described a kind of electric load of claim 1, it is characterized in that described D step sets up the multidimensional phase space of electric load chaos time sequence, when formation learning sample and teacher are worth, determine optimum delay time with the autocorrelation function method τ, determine the best dimension that embeds with saturated correlation dimension method m, the phase space of reconfiguration system:
Figure 2012103607943100001DEST_PATH_IMAGE002
Wherein N=n-( M-1) * τLength for sequence vector.
3. the chaos phase space neural network model safety monitoring method of the described a kind of electric load of claim 1 when it is characterized in that described E step is determined neural network structure, by the input layer dimension of network is m, i.e. the embedding dimension of chaos phase space, output dimension are one dimension (i.e. the output of prediction), the Hidden unit dimension is determined in network training study by trial and error.
4. the chaos phase space neural network model safety monitoring method of the described a kind of electric load of claim 1 when it is characterized in that the normalized of described F step data, with unified method for normalizing, is used conversion to all sample values, teacher's value and output valve
Figure 2012103607943100001DEST_PATH_IMAGE003
Load is scaled [0,1] interval value.
5. the chaos phase space neural network model safety monitoring method of the described a kind of electric load of claim 1 when it is characterized in that the study of described G step network, sequentially comprises following characteristics:
Any two phase points in the chaos phase space
Figure 2012103607943100001DEST_PATH_IMAGE004
With
Figure 2012103607943100001DEST_PATH_IMAGE005
(wherein TBe leading time), their nonlinear Evolution funtcional relationships in phase space
Figure 2012103607943100001DEST_PATH_IMAGE006
, come match by artificial neural network;
Although prediction output point
Figure 161615DEST_PATH_IMAGE005
For mN dimensional vector n, but in fact ( m-1) dimension is known quantity, gets for the sake of simplicity network and is output as
Figure 511825DEST_PATH_IMAGE005
mIndividual component;
To the neuron j of any one non-linear unit, its input/output relation has:
（1）
D. in the formula, xBe neuronic input, ω is connection weight, and θ is threshold value, yBe this neuronic output;
E. f is excitation function, commonly used is the Sigmoid function, that is:
（2）
F. error backpropagation algorithm (BP algorithm) is pressed in the study of the connection weight of neural network and threshold value, and this is a kind of teacher's of having learning algorithm;
G. learning process comprises positive and negative two processes, at first by the output of each layer of formula (1) forward computational grid, calculate again the error between output and the teacher value, utilize error oppositely to revise connection weight and threshold value, repeat above two processes until the error of output and sample teacher value is little of a certain setting value ε;
H. hypothesis input learning sample is kIt is individual,
Figure 2012103607943100001DEST_PATH_IMAGE009
, the predicted value of output is
Figure 2012103607943100001DEST_PATH_IMAGE010
, corresponding teacher's value is
Figure 2012103607943100001DEST_PATH_IMAGE011
, objective function is:
Figure 2012103607943100001DEST_PATH_IMAGE012
（3）
I. gradient descent method is adopted in the correction of connection weight and threshold value, in order to prevent concussion and accelerating convergence, adopts the algorithm of additional momentum item:
Figure DEST_PATH_IMAGE013
（4）
In the formula, Connection weight or the threshold value of difference input layer, hidden layer and output layer,
Figure DEST_PATH_IMAGE015
Be iterations, η is training speed,
Figure 2012103607943100001DEST_PATH_IMAGE016
Be the modification value,
Figure DEST_PATH_IMAGE017
Be factor of momentum.
6. the chaos phase space neural network model safety monitoring method of the described a kind of electric load of claim 1 reports phase point to be input to neural network rising of chaos phase space when it is characterized in that described H step forecast model, and the output of network is predicted value.
CN2012103607943A 2012-09-21 2012-09-21 Safety monitoring method for neural network model of power-loaded chaotic phase space Pending CN102915511A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2012103607943A CN102915511A (en) 2012-09-21 2012-09-21 Safety monitoring method for neural network model of power-loaded chaotic phase space

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2012103607943A CN102915511A (en) 2012-09-21 2012-09-21 Safety monitoring method for neural network model of power-loaded chaotic phase space

Publications (1)

Publication Number Publication Date
CN102915511A true CN102915511A (en) 2013-02-06

Family

ID=47613866

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2012103607943A Pending CN102915511A (en) 2012-09-21 2012-09-21 Safety monitoring method for neural network model of power-loaded chaotic phase space

Country Status (1)

Country Link
CN (1) CN102915511A (en)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103268519A (en) * 2013-04-26 2013-08-28 哈尔滨工程大学 Electric power system short-term load forecast method and device based on improved Lyapunov exponent
CN106845863A (en) * 2017-02-23 2017-06-13 沈阳工业大学 A kind of distributed wind-power generator is exerted oneself and heat load sync index Forecasting Methodology
CN106897796A (en) * 2017-02-23 2017-06-27 沈阳工业大学 Distributed light stores up generated output to operation of air conditioner stability influence index forecasting method
CN107153870A (en) * 2017-05-12 2017-09-12 沈阳工程学院 The power prediction system of small blower fan
CN108510072A (en) * 2018-03-13 2018-09-07 浙江省水文局 A kind of discharge of river monitoring data method of quality control based on chaotic neural network
CN108564201A (en) * 2018-03-16 2018-09-21 电子科技大学 A kind of close interval prediction method of salt based on phase space reconfiguration and quantile estimate
CN112232593A (en) * 2020-11-04 2021-01-15 武汉理工大学 Power load prediction method based on phase space reconstruction and data driving
CN112532643A (en) * 2020-12-07 2021-03-19 长春工程学院 Deep learning-based traffic anomaly detection method, system, terminal and medium
CN113203471A (en) * 2021-05-07 2021-08-03 国网山西省电力公司电力科学研究院 Transformer mechanical fault detection method based on wavelet neural network prediction
CN114282639A (en) * 2021-12-24 2022-04-05 上海应用技术大学 Water bloom early warning method based on chaos theory and BP neural network
CN116415510A (en) * 2023-06-12 2023-07-11 国网江西省电力有限公司电力科学研究院 Breaker temperature rise prediction method and system based on phase space reconstruction and neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李眉眉等: "基于混沌分析的BP 神经网络模型及其在负荷预测中的应用", 《四川大学学报（工程科学版）》 *

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103268519A (en) * 2013-04-26 2013-08-28 哈尔滨工程大学 Electric power system short-term load forecast method and device based on improved Lyapunov exponent
CN103268519B (en) * 2013-04-26 2016-05-25 哈尔滨工程大学 Based on the power-system short-term load forecasting method and the device that improve Lyapunov index
CN106845863A (en) * 2017-02-23 2017-06-13 沈阳工业大学 A kind of distributed wind-power generator is exerted oneself and heat load sync index Forecasting Methodology
CN106897796A (en) * 2017-02-23 2017-06-27 沈阳工业大学 Distributed light stores up generated output to operation of air conditioner stability influence index forecasting method
CN107153870A (en) * 2017-05-12 2017-09-12 沈阳工程学院 The power prediction system of small blower fan
CN108510072A (en) * 2018-03-13 2018-09-07 浙江省水文局 A kind of discharge of river monitoring data method of quality control based on chaotic neural network
CN108564201A (en) * 2018-03-16 2018-09-21 电子科技大学 A kind of close interval prediction method of salt based on phase space reconfiguration and quantile estimate
CN112232593A (en) * 2020-11-04 2021-01-15 武汉理工大学 Power load prediction method based on phase space reconstruction and data driving
CN112532643A (en) * 2020-12-07 2021-03-19 长春工程学院 Deep learning-based traffic anomaly detection method, system, terminal and medium
CN112532643B (en) * 2020-12-07 2024-02-20 长春工程学院 Flow anomaly detection method, system, terminal and medium based on deep learning
CN113203471A (en) * 2021-05-07 2021-08-03 国网山西省电力公司电力科学研究院 Transformer mechanical fault detection method based on wavelet neural network prediction
CN114282639A (en) * 2021-12-24 2022-04-05 上海应用技术大学 Water bloom early warning method based on chaos theory and BP neural network
CN114282639B (en) * 2021-12-24 2024-02-02 上海应用技术大学 Water bloom early warning method based on chaos theory and BP neural network
CN116415510A (en) * 2023-06-12 2023-07-11 国网江西省电力有限公司电力科学研究院 Breaker temperature rise prediction method and system based on phase space reconstruction and neural network
CN116415510B (en) * 2023-06-12 2023-09-12 国网江西省电力有限公司电力科学研究院 Breaker temperature rise prediction method and system based on phase space reconstruction and neural network

Similar Documents

Publication Publication Date Title
CN102915511A (en) Safety monitoring method for neural network model of power-loaded chaotic phase space
Liu et al. Forecasting power output of photovoltaic system using a BP network method
CN109255505B (en) Short-term load prediction method of multi-model fusion neural network
Guo et al. Multi-step forecasting for wind speed using a modified EMD-based artificial neural network model
CN102930358B (en) A kind of neural net prediction method of photovoltaic power station power generation power
CN104951836A (en) Posting predication system based on nerual network technique
CN106875041A (en) A kind of short-term wind speed forecasting method
CN103218674A (en) Method for predicating output power of photovoltaic power generation system based on BP (Back Propagation) neural network model
CN104978611A (en) Neural network photovoltaic power generation output prediction method based on grey correlation analysis
CN105160423A (en) Photovoltaic power generation prediction method based on Markov residual error correction
CN103927460A (en) Wind power plant short-term wind speed prediction method based on RBF
CN103793887A (en) Short-term electrical load on-line predicting method based on self-adaptation enhancing algorithm
CN104820877A (en) Photovoltaic system generation power prediction method based on cloud adaptive PSO-SNN
CN103425867A (en) Short-term wind power combination prediction method
Lodge et al. Short term wind speed prediction using artificial neural networks
Patel et al. Short term load forecasting of Indian system using linear regression and artificial neural network
CN111489038A (en) Method for improving accuracy of ultra-short-term power prediction of wind power plant
Mukherjee et al. Solar irradiance prediction from historical trends using deep neural networks
He et al. Short-term power load forecasting based on self-adapting PSO-BP neural network model
Zhong et al. PV power short-term forecasting model based on the data gathered from monitoring network
Arshad et al. Wind power prediction using genetic programming based ensemble of artificial neural networks (GPeANN)
Li et al. A short-term power load forecasting method based on BP neural network
Paulin et al. SOLAR PHOTOVOLTAIC OUTPUT POWER FORECASTING USING BACK PROPAGATION NEURAL NETWORK.
Datta An artificial neural network approach for short-term wind speed forecast
Zhenhao et al. Prediction of wind power ramp events based on deep neural network

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20130206