CN108663570B - Current harmonic analysis method based on trigonometric function neural network - Google Patents
Current harmonic analysis method based on trigonometric function neural network Download PDFInfo
- Publication number
- CN108663570B CN108663570B CN201810222379.9A CN201810222379A CN108663570B CN 108663570 B CN108663570 B CN 108663570B CN 201810222379 A CN201810222379 A CN 201810222379A CN 108663570 B CN108663570 B CN 108663570B
- Authority
- CN
- China
- Prior art keywords
- neural network
- current
- trigonometric function
- harmonic
- representing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- 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
Abstract
The invention relates to a current harmonic analysis method based on a trigonometric function neural network, which aims at harmonic generated by electric vehicle charging to a power system and realizes harmonic analysis of electric vehicle charging current. Converting the charging current of the electric automobile into a trigonometric function expressed by weight, constructing a trigonometric function neural network, and acquiring output current containing each harmonic component through forward recursion of the trigonometric function neural network; comparing the output current with the input current, and obtaining the optimal weight of the trigonometric function neural network by adopting a negative gradient descent method through the reverse iteration of the trigonometric function neural network by the difference value of the output current and the input current, thereby obtaining the accurate estimation of the charging current harmonic parameters of the electric automobile. The method has faster convergence characteristics and better noise tolerance.
Description
Technical Field
The invention relates to a power detection technology, in particular to a current harmonic analysis method based on a trigonometric function neural network.
Background
Conventional chargers suffer from low-order, low power factor, uncontrollable state of charge, high input current harmonics, and limited control of battery current. Moreover, the low frequency and low power factor high input current harmonics do not meet IEC1000-3-2 or IEEE519 harmonic standards, and the uncontrolled state of charge will shorten battery life.
To mitigate harmonic pollution, accurate monitoring and analysis of harmonic components is paramount, and compensation techniques are applied to correct waveforms, or terminate power transmission. Therefore, to enhance harmonic monitoring, estimation of harmonic parameters such as magnitude and phase is particularly important. To date, the field of harmonic measurement of power systems has been greatly advanced, wherein the most common method for estimating harmonic components is based on Fast Fourier Transform (FFT), but due to the phenomenon of sharp stake fence and frequency spectrum leakage, FFT has certain limitations in practical application of harmonic analysis of charging current of a charging car, and generally causes estimation errors of amplitude, phase, frequency and the like of harmonic current. Artificial Neural Networks (ANNs) have attracted widespread attention over the last two decades, and the organic combination of ANNs with harmonics to achieve accurate estimation of the harmonics has become a focus of attention, because of its computational speed and robustness, to be able to track the individual components of the current harmonics quickly and accurately. However, how to combine the charging current of the electric vehicle with the artificial neural network and accurately calculate the amplitude and phase of the harmonic current is a great difficulty at present.
Disclosure of Invention
Aiming at the problem that the self fault or abnormality of the electric vehicle charger affects the power quality of a power system, the invention provides a current harmonic analysis method based on a trigonometric function neural network, which is used for carrying out harmonic analysis on charging current of an electric vehicle and can rapidly realize analysis on parameters of each subharmonic of the current.
The technical scheme of the invention is as follows: the current harmonic analysis method based on the trigonometric function neural network converts the charging current of the electric automobile into a trigonometric function expressed by weight, constructs the trigonometric function neural network, and obtains the accurate estimation of the charging current harmonic parameter of the electric automobile by solving the optimal weight of the trigonometric function neural network, and specifically comprises the following steps:
1) Constructing trigonometric function neural network
The charging current of the electric automobile is expressed as a trigonometric function:
wherein: i (t) represents a charging current of the electric vehicle; i.e dc (t) represents a direct current component of an electric vehicle charging current; n represents the harmonic frequency of charging current of the electric automobile; omega 0 Angular frequency w representing fundamental component j Represents the jth current harmonic parameter, w N+j Representing the n+j current harmonic parameter;
the neural network input layer is n pairs of training data sets { x, y }, x, y epsilon R 1*n A matrix, where x= (x (1), x (2), … …, x (i), … … x (n)), y= (y (1), y (2), … …, y (i), … … y (n)), x (i) =t (i) is the time corresponding to the i-th sampling point; y (i) is the instantaneous value of the current corresponding to the ith sampling point;
trigonometric function for neural network hidden layer
Constructing an activation function, and totally 2N+1Hidden neurons, designing activation function of hidden layer +.>
The method comprises the following steps:
for the output layer of neural network
The representation is:
wherein:
representing the output current represented by the trigonometric function neural network corresponding to the ith sampling time; w (w) 0 Representing weights between hidden layer 1 st neurons and output layer; w (w) j Representing the jth current harmonic parameter and also representing the weight between the jth neuron of the hidden layer and the output layer; w (w) N+j Representing the n+j current harmonic parameter and also representing the weight between the n+j neuron of the hidden layer and the output layer;
2) Updating weights by using a performance function and a negative gradient descent method to solve and analyze current harmonics:
the design learning rule performance function is:
comparing the calculated and estimated output current of the neural network with the input current, when the target error is not satisfied, i.e. e (w) does not reach the desired set value e obj When the weight is updated by a negative gradient iteration method;
based on a negative gradient descent method, designing a weight iteration formula of the trigonometric function neural network comprises the following steps:
w(k+1)＝w(k)-ηP T (Pw(k)-y)
wherein:
wherein: k=1, 2, …, iter max Represents the kth iteration, iter max The iteration times; w (k) represents the weight corresponding to the kth iteration; w (k+1) represents the weight corresponding to the k+1st iteration; η represents a learning rate; if eta is more than 0 and is small enough, the weight of the trigonometric function neural network can be converged to the optimal weight through iteration;
by solving the optimal weight vector of the neural network, the amplitude and phase angle of the direct current component and the jth harmonic can be estimated through a weight iteration formula:
wherein: a is that dc Representing the magnitude of the direct current component; a is that j Representing the amplitude of the jth harmonic; phi (phi) j Representing the phase of the jth harmonic.
The invention has the beneficial effects that: according to the current harmonic analysis method based on the trigonometric function neural network, the weight is updated in a negative gradient descending mode, so that accurate harmonic estimation of current or voltage is obtained, and harmonic components can be effectively calculated. The method provides a simple and feasible solution for harmonic estimation.
Drawings
FIG. 1 is a block diagram of a trigonometric function based neural network of the present invention;
FIG. 2 is a flow chart of a harmonic analysis method based on a trigonometric function neural network according to the present invention;
FIG. 3a is a graph showing the comparison of the performance of the trigonometric function neural network and FFT of the present invention in analyzing the magnitude of the fundamental wave of the charging current of an electric vehicle;
FIG. 3b is a graph showing the comparison of the performance of the trigonometric function neural network and FFT of the present invention in analyzing the magnitude of the third harmonic of the charging current of an electric vehicle;
fig. 4 is a graph showing the performance comparison of the trigonometric function neural network and the FFT in analyzing the fundamental phase angle of the charging current of the electric vehicle.
Detailed Description
The current harmonic analysis method based on the trigonometric function neural network comprises the following steps: the method mainly comprises the steps of converting charging current of an electric automobile into a trigonometric function expressed by weight, constructing a trigonometric function neural network, and obtaining output current containing harmonic components through forward recursion of the trigonometric function neural network; comparing the output current with the input current, the difference value of the output current and the input current adopts a negative gradient descent method to calculate the optimal weight of the trigonometric function neural network through the reverse iteration of the trigonometric function neural network, and the optimal weight can be used for rapidly and accurately estimating the harmonic component and related parameters of the system, so that the robustness is good, and the analysis of the current subharmonic parameters can be rapidly realized.
The electric vehicle charging current can be expressed in terms of amplitude and phase as the sum of the individual harmonic components:
wherein: i (t) represents a charging current of the electric vehicle; i.e dc (t) represents a direct current component of an electric vehicle charging current; n represents the harmonic frequency of charging current of the electric automobile; omega 0 An angular frequency representing the fundamental component; a is that j Representing the amplitude of the jth harmonic; phi (phi) j Representing the phase of the jth harmonic.
According to the triangle equation, define
φ j ＝tan -1 [w j /w N+j ]Then formula (1) can be written as:
wherein: w (w) j Represents the jth current harmonic parameter, w N+j Indicating the N + j current harmonic parameter.
As shown in FIG. 1, the input layer is n pairs of training data sets { x, y }, x, y ε R 1*n A matrix. Where x= (x (1), x (2), … …, x (i), … … (n)), y= (y (1), y (2), … …, y (i), … … y (n)), x (i) =t (i) is the time corresponding to the i-th sampling point; y (i) is the instantaneous value of the current corresponding to the ith sample point.
Trigonometric function for hidden layer
Constructing an activation function, totaling 2N+1 hidden neurons, designing the activation function of the hidden layer +.>
The method comprises the following steps:
for the output layer
The representation is:
wherein:
representing the output current represented by the trigonometric function neural network corresponding to the ith sampling time; w (w) 0 Representing weights between hidden layer 1 st neurons and output layer; w (w) j Representing the jth current harmonic parameter and also representing the weight between the jth neuron of the hidden layer and the output layer; w (w) N+j The n+j current harmonic parameter is represented, as is the weight between the n+j neuron of the hidden layer and the output layer.
The design learning rule performance function is:
based on a negative gradient descent method, designing a weight iteration formula of the trigonometric function neural network comprises the following steps:
w(k+1)＝w(k)-ηP T (Pw(k)-y) (6)
wherein:
wherein: k=1, 2, …, iter max Represents the kth iteration, iter max The iteration times; w (k) represents the weight corresponding to the kth iteration; w (k+1) represents the weight corresponding to the k+1st iteration; η represents the learning rate. If η > 0 and is small enough, the weights of the trigonometric neural network may be converged to the optimal weights by iteration.
By solving the optimal weight vector of the neural network, the amplitude and phase angle of the direct current component and the jth harmonic can be estimated by the equation (6):
wherein: a is that dc Representing the magnitude of the direct current component; a is that j Representing the amplitude of the jth harmonic; phi (phi) j Representing the phase of the jth harmonic.
The flow chart of the harmonic analysis method based on the trigonometric function neural network is shown as figure 2, the output current and the input current of the neural network obtained by calculation and estimation are compared, when the target error is not satisfied, namely e (w) does not reach the desired set value e obj And when the current is estimated, the weight is updated by a negative gradient iteration method, so that the actual current is estimated accurately and rapidly.
In order to verify the correctness and effectiveness of the trigonometric function neural network analysis harmonic wave, simulation research is carried out on the current signal actually containing the harmonic wave through MATLAB/Simulink. The input data is the actual value of the power system current and the various harmonic components of the current signal under consideration are shown in table one. The input current signal is calculated by table 1 and equation (1).
TABLE 1
| Harmonic order | Amplitude value | Phase of | Order of the | Amplitude value | Phase of |
| 1 | 3.00 | -23.10 | 6 | 0.00 | - |
| 2 | 0.03 | 115.60 | 7 | 003 | -31.80 |
| 3 | 0.15 | 59.30 | 8 | 0.00 | - |
| 4 | 0.01 | 52.40 | 9 | 0.01 | -63.70 |
| 5 | 0.04 | 123.80 | 10 | - | - |
Omega during simulation 0 Let 100 pi rad/s, learning rate eta of 0.01, harmonic maximum number N=9, sampling at sampling frequency f=1000 Hz per iteration, 10 points total (n=10), target error e obj Set to 0.01.
To prove the superiority of the trigonometric function neural network in analyzing the harmonic wave, the trigonometric function neural network is compared with the Fourier decomposition (FFT) simulation, and the simulation results are shown in figures 3a, 3b and 4. Both the ANN and the FFT can track the amplitude and the phase angle of the fundamental wave and other components, the FFT needs at least one period of 0.02s to calculate the corresponding amplitude and phase angle, and the ANN can track the corresponding amplitude and phase angle only by half period of 0.01s, which is beneficial to real-time monitoring and rapid protection. The ANN is able to track the individual components of the actual current signal quickly compared to the waveform under FFT control.
Claims (1)
1. A current harmonic analysis method based on trigonometric function neural network is characterized in that,
converting the charging current of the electric automobile into a trigonometric function expressed by weight, constructing a trigonometric function neural network, and obtaining accurate estimation of the harmonic parameters of the charging current of the electric automobile by solving the optimal weight of the trigonometric function neural network, wherein the method specifically comprises the following steps:
1) Constructing trigonometric function neural network
The charging current of the electric automobile is expressed as a trigonometric function:
wherein: i (t) represents a charging current of the electric vehicle; i.e dc (t) represents a direct current component of an electric vehicle charging current; n represents the harmonic frequency of charging current of the electric automobile; omega 0 An angular frequency representing the fundamental component; w (w) j Represents the jth current harmonic parameter, w N+j Representing the n+j current harmonic parameter;
the neural network input layer is n pairs of training data sets { x, y }, x, y epsilon R 1*n A matrix, where x= (x (1), x (2), … …, x (i), … … x (n)), y= (y (1), y (2), … …, y (i), … … y (n)), x (i) =t (i) is the time corresponding to the i-th sampling point; y (i) is the instantaneous value of the input current corresponding to the ith sampling point;
trigonometric function for neural network hidden layer
Constructing an activation function, totaling 2N+1 hidden neurons, designing the activation function of the hidden layer +.>
The method comprises the following steps:
for the output layer of neural network
The representation is:
wherein:
representing the output current represented by the trigonometric function neural network corresponding to the ith sampling time; w (w) 0 Representing weights between hidden layer 1 st neurons and output layer; w (w) j Representing the jth current harmonic parameter and also representing the weight between the jth neuron of the hidden layer and the output layer; w (w) N+j Representing the n+j current harmonic parameter and also representing the weight between the n+j neuron of the hidden layer and the output layer;
2) Updating weights by using a performance function and a negative gradient descent method to solve and analyze current harmonics:
the design learning rule performance function is:
comparing the calculated and estimated output current of the neural network with the instantaneous value of the input current, when the target error is not satisfied, i.e. e (w) reaches the undesirable set value e obj When the weight is updated by a negative gradient iteration method;
based on a negative gradient descent method, designing a weight iteration formula of the trigonometric function neural network comprises the following steps:
w(k+1)＝w(k)-ηP T (Pw(k)-y)
wherein:
wherein: k=1, 2, …, iter max Represents the kth iteration, iter max The iteration times; w (k) represents the weight corresponding to the kth iteration; w (k+1) represents the weight corresponding to the k+1st iteration; η represents a learning rate; if let eta>0 and small enough, the weight of the trigonometric function neural network can be converged to the optimal weight through iteration;
by solving the optimal weight vector of the neural network, the amplitude and phase angle of the direct current component and the jth harmonic can be estimated through a weight iteration formula:
wherein: a is that dc Representing the magnitude of the direct current component; a is that j Representing the amplitude of the jth harmonic; phi (phi) j Representing the phase of the jth harmonic.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810222379.9A CN108663570B (en) | 2018-03-15 | 2018-03-15 | Current harmonic analysis method based on trigonometric function neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810222379.9A CN108663570B (en) | 2018-03-15 | 2018-03-15 | Current harmonic analysis method based on trigonometric function neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108663570A CN108663570A (en) | 2018-10-16 |
| CN108663570B true CN108663570B (en) | 2023-05-23 |
Family
ID=63785241
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810222379.9A Active CN108663570B (en) | 2018-03-15 | 2018-03-15 | Current harmonic analysis method based on trigonometric function neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108663570B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800520B (en) * | 2019-01-25 | 2023-03-24 | 国网浙江省电力有限公司湖州供电公司 | Electric vehicle charging station harmonic modeling method based on neural network |
| CN112117947B (en) * | 2020-09-30 | 2022-03-11 | 桂林电子科技大学 | SRM torque ripple suppression control system and method based on current injection method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101216512A (en) * | 2007-12-29 | 2008-07-09 | 湖南大学 | Non-sine periodic signal real time high precision detection method |
| CN102338827A (en) * | 2011-06-10 | 2012-02-01 | 中国矿业大学 | Method for analyzing electric network waveform distortions and automatically monitoring electric power harmonic parameters |
| CN102353839A (en) * | 2011-07-18 | 2012-02-15 | 华北电力大学（保定） | Electric power system harmonics analysis method based on multilayered feedforward neural network |
| CN102998527A (en) * | 2012-11-26 | 2013-03-27 | 上海电力学院 | Pass band type fundamental wave, harmonic wave and direct current component detection method |
| CN103383413A (en) * | 2013-07-09 | 2013-11-06 | 温州大学 | Real-time harmonic detection method based on direct weight determination method |
| CN103424621A (en) * | 2013-08-20 | 2013-12-04 | 江苏大学 | Artificial neural network detecting method of harmonic current |
| CN104777356A (en) * | 2015-03-10 | 2015-07-15 | 三峡大学 | Neural-network-based real-time high-accuracy harmonic detection method |
-
2018
- 2018-03-15 CN CN201810222379.9A patent/CN108663570B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101216512A (en) * | 2007-12-29 | 2008-07-09 | 湖南大学 | Non-sine periodic signal real time high precision detection method |
| CN102338827A (en) * | 2011-06-10 | 2012-02-01 | 中国矿业大学 | Method for analyzing electric network waveform distortions and automatically monitoring electric power harmonic parameters |
| CN102353839A (en) * | 2011-07-18 | 2012-02-15 | 华北电力大学（保定） | Electric power system harmonics analysis method based on multilayered feedforward neural network |
| CN102998527A (en) * | 2012-11-26 | 2013-03-27 | 上海电力学院 | Pass band type fundamental wave, harmonic wave and direct current component detection method |
| CN103383413A (en) * | 2013-07-09 | 2013-11-06 | 温州大学 | Real-time harmonic detection method based on direct weight determination method |
| CN103424621A (en) * | 2013-08-20 | 2013-12-04 | 江苏大学 | Artificial neural network detecting method of harmonic current |
| CN104777356A (en) * | 2015-03-10 | 2015-07-15 | 三峡大学 | Neural-network-based real-time high-accuracy harmonic detection method |
Non-Patent Citations (3)
| Title |
|---|
| 《基于三角函数神经网络的电动汽车充电电流谐波分析方法》;陈忠华 等;《上海电力学院学报》;20181215;全文 * |
| 基于分段迭代的电力谐波神经网络分析方法;操吴兵等;《电测与仪表》;20141210(第23期);全文 * |
| 基于神经网络的高精度电力系统谐波分析;雷锡社等;《电测与仪表》;20070325(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108663570A (en) | 2018-10-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108155648B (en) | State estimation method based on adaptive H-infinity extended Kalman filtering | |
| Vafaeipour et al. | Application of sliding window technique for prediction of wind velocity time series | |
| CN109001640B (en) | Data processing method and device for power battery | |
| CN105974327A (en) | Lithium battery pack SOC prediction method based on neural network and UKF | |
| Ao et al. | Proton exchange membrane fuel cell prognosis based on frequency-domain Kalman filter | |
| CN110991721A (en) | Short-term wind speed prediction method based on improved empirical mode decomposition and support vector machine | |
| CN109444505B (en) | Harmonic current detection algorithm for electric vehicle charging station | |
| CN108663570B (en) | Current harmonic analysis method based on trigonometric function neural network | |
| Alao et al. | Online condition monitoring of sealed lead acid & lithium nickel-cobalt-manganese oxide batteries using broadband impedance spectroscopy | |
| Ye et al. | An online model-based battery parameter and state estimation method using multi-scale dual adaptive particle filters | |
| Mukhopadhyay et al. | Supercapacitor characterization using universal adaptive stabilization and optimization | |
| CN110210152B (en) | Ultrahigh-order harmonic source modeling method | |
| CN110163148A (en) | A kind of electric car DC charging distorted signal self-adaptive identification method | |
| Cai et al. | Wind power forecasting based on ensemble empirical mode decomposition with generalized regression neural network based on cross-validated method | |
| Dash et al. | Accurate tracking of harmonic signals in VSC-HVDC systems using PSO based unscented transformation | |
| Qin et al. | State of Charge estimation of lithium-ion power battery based on online parameter identification method and BP neural network | |
| Saoudi et al. | State estimation of unknown input fuzzy bilinear systems: Application to fault diagnosis | |
| CN109167395B (en) | Transient equivalent potential identification method of photovoltaic power generation system based on ADPSS | |
| Xue et al. | Remaining useful life estimation for proton exchange membrane fuel cell based on extreme learning machine | |
| Wang et al. | A wind speed prediction method based on improved empirical mode decomposition and support vector machine | |
| Li et al. | A new method to estimate the state of charge of the green battery | |
| Kuang et al. | Real-time forecasting for short-term traffic flow based on general regression neural network | |
| Shen et al. | A novel neural network approach for power system low frequency oscillation mode identification | |
| Guoqiang et al. | Study of RBF neural network based on PSO algorithm in nonlinear system identification | |
| CN112448392B (en) | Regional power grid harmonic treatment method and device based on complex-valued neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Yu Rongjiang Inventor after: Wang Xinyue Inventor after: Zhang Fan Inventor after: Jin Jiao Inventor after: Zhu Yijia Inventor after: Shen Guoheng Inventor after: Luo Jinsheng Inventor after: Chen Zhonghua Inventor after: Wang Yufei Inventor after: Hu Chengang Inventor after: Chen Pan Inventor after: Xue Hua Inventor after: Xu Xiuzhen Inventor after: Chen Bing Inventor before: Yu Rongjiang Inventor before: Wang Xinyue Inventor before: Zhang Fan Inventor before: Jin Jiao Inventor before: Zhu Yijia Inventor before: Shen Guoheng Inventor before: Luo Jinsheng Inventor before: Chen Zhonghua Inventor before: Wang Yufei Inventor before: Hu Chengang Inventor before: Chen Pan Inventor before: Xue Hua Inventor before: Xu Xiuzhen Inventor before: Chen Bing |
|
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |