Journal of Electrical Engineering and Electronic TechnologyISSN: 2325-9833

All submissions of the EM system will be redirected to Online Manuscript Submission System. Authors are requested to submit articles directly to Online Manuscript Submission System of respective journal.

Review Article, J Electr Eng Electron Technol Vol: 7 Issue: 1

Photovoltaic Power Tracking Techniques with Intelligent Prediction

Baig AA*, Yang XX and Yang MS

College of Electrical and Information Technology, Changsha University of Science and Technology, China

*Corresponding Author : Baig Ammad Anwar
College of Electrical and Information Technology, Changsha University of science and technology Changsha, Hunan-410004, China
Tel: +(86) 13787169647
[email protected]

Received: December 18, 2017 Accepted: January 23, 2018 Published: February 01, 2018

Citation: Baig AA, Yang XX , Yang MS (2018) Photovoltaic Power Tracking Techniques with Intelligent Prediction. J Electr Eng Electron Technol 7:1. doi: 10.4172/2325-9833.1000154


Prediction of photovoltaic (PV) power output has become more important recently. PV power completely depends on solar cell working conditions, which are directly connected to the weather. So, to get the maximum power output from the system many maximum power point tracking (MPPT) techniques have been introduced, but in this paper, MPPT techniques with intelligent algorithm have been discussed. These techniques are mathematically modeled and presented in such a way that the appropriate application could be chosen. A comparative table is presented at the end of the paper to simplify the classification of the different techniques.

Keywords: Intelligent power prediction; Maximum power point tracking; Photovoltaic generation


Prediction of photovoltaic (PV) power output has become more important recently. PV power is an inexhaustible and broadly available energy resource. However, the output power of photovoltaic modules depends on solar radiation and the temperature of the solar cells. Despite being cost efficient, the efficiency of PV system is less than 20%, and that is only in a typical sunny day. If it is going to be a rainy or cloudy day, power output from the photovoltaic system drops dramatically. Since PV system completely depends on weather conditions researchers are eager to find new prediction techniques to get maximum power out of PV systems in suboptimal conditions. Successful power prediction of a PV system helps dispatching the power to the grid with improved efficiency. For this, intelligent tools are needed to predict the real power output of PV, and to track the maximum power point of the PV array.

In this paper, maximum power point tracking (MPPT) techniques with intelligent algorithm have been analyzed, which include fuzzy logic control technique, biological swarm chasing technique, PCA and SVM technique, weighted gaussian process regression (WGPR) technique, support vector regression (SVR) technique, auto-regressive integrated moving average (ARIMA) technique, artificial neural network (ANN) and extreme learning machine with forgetting mechanism FOS-ELM technique.

Comparison of Techniques

Technique 1: Fuzzy logic control

Trackers based on Fuzzy calculation are considered smart, since they track the MPP even if the inputs are imprecise. Fuzzy controllers do not need an accurate mathematical model. In general, fuzzy control consists of three stages, called fuzzification, rule base lookup table and defuzzification. In the first modeling stage, the numerical input variables are converted into linguistic variables based on a membership function, where five fuzzy levels are used: NB (negative big), NS (negative small), ZE (zero), PS (positive small) and PB (positive big).

The inputs to an MPPT fuzzy logic controller are usually an error E and a change in error ΔE expressed as:

image (1)


image (2)

Once E and ΔD are calculated and converted to the linguistic variables, the output of the fuzzy controller, which is the change in duty-cycle ΔD of the power converter, can be found in the table shown in Figure 1. In the defuzzification stage, the fuzzy logic controller output is converted from a linguistic variable to a numerical variable and provides an analog signal that will drive the power converter to the MPP. The MPPT fuzzy logic controller works well under varying atmospheric conditions. However, its effectiveness depends on choosing the right error computation and coming up with the rule base table [1,2].

Figure 1: Fuzzy rule table.

Technique 2: Biological swarm chasing algorithm

A smart technique was proposed by Chen et al. [3] for tracking the MPP in the same way as the Particle Swarm Optimization (PSO) technique. Swarm intelligence is an artificial intelligence technique involving the study of collective behavior in decentralized systems. One of the most popular swarm intelligence paradigms is the (PSO), which is basically developed through the simulation of the social behavior of bird flocking and fish schooling. In the proposed Bio-MPPT-based PV power system, each PV module is viewed as a particle, and the MPP is viewed as the moving target. Every PV module is designed to chase the MPP automatically by using the Bio-MPPT algorithm. Thus, by applying this method to a PV system, all modules in the PV array are assumed slaves for only one master module and each module has its own controller that communicates with the master controller to achieve the chasing of the MPP. Chen et al. [3] showed an efficiency enhancement of 12.19% compared to the P & O method.

Technique 3: PCA and SVM

Principle component analysis (PCA) and support vector machine (SVM) is a novel forecasting technique. It is the combination between Principal Component Analysis (PCA) and Support Vector Machine (SVM) intelligent algorithm. PCA statistical method used to extract less principal components instead of the original meteorological factor. Then that is as the input of the SVM model. The prerequisite is to have a strong correlation between the numbers of the original variables before using PCA. If the original relationship between variables is weak, it indicate that variables have a lower degree of redundancy, which will not be able to extract less common factor from the comprehensive variable to reflect the common characteristics. Therefore, the original variable needed to be determined for relation with each other. KMO test is the index of the correlation coefficient and partial correlation coefficients between the comparative correlation variable. Mathematically defined as:

image (3)

Where rij is the simple correlation coefficient between variables xi and other variables xj, pij is the partial correlation coefficients between variable xi and other variables xj in the control of the remaining variables. KMO statistic value is between 0~1. KMO value is closer to 1, which illustrates the correlation coefficient is strong. When there is strong correlation between the original variable, more suitable it is for PCA analysis.

In this technique, model forecasting can not only improve the prediction accuracy of the photovoltaic power output also reduce the computational model to meet real time requirements for forecasting. More meteorological variables mainly caused by the mutual influence among some of the meteorological variables, because they can affect the predictive PV output. But with the PCS-SVM technique, model prediction error is less than 14% [4].

Technique 4: Weighted Gaussian process regression

Weighted Gaussian process regression (WGPR) approach is proposed on samples with higher outlier potential that has low weight. A density-based local outlier detection approach is introduced to compensate the deterioration of Euclidean distance for high-dimensional data. A novel concept of the degree of nonlinear correlation is incorporated to compute the contribution of every individual data attribute.

In data modeling, the points with larger dispersion have more influence on the model. To avoid the model deviation caused by heteroscedasticity, the outliers are assigned with lower weight. WGPR is based on GPR, which is derived from function space view and weight space view. This model technique is complex but author in this paper presented a method with an efficient solution for extracting information from high-dimensional data. In addition, Satellite images, sky images, and other weather related data can be introduced in the future work for more comprehensive weather information. For applications with higher real-time requirements, feature extraction approaches can be combined to improve better operational efficiency. In this technique, model prediction nRMSE error is less than 6.646, 6.647% in GPR and WGPR respectively in mid of December (Figure 2)[5].

Figure 2: Flow chart of WGPR forecasting process.

Technique 5: Support vector regression

SVM (support vector machine) is a nonparametric technique for data classification and regression [6], developed within the framework of statistical learning theory or VC (Vapnik–Chervonenkis) theory that studies properties of learning machines [6]. A Support Vector Regression (SVR) technique is used for forecasting. Three models ε-SVR, ν-SVR and LS-SVR, which are based on SVR are compared using performance indicator, nRMSE [8].

ε-SVR model function f actually exists all pairs with ε precision. But it is difficult to select an appropriate ε because of its complexity.

ν-SVR: In [7], a new parameter ν is introduced which allows controlling the number of support vectors and training errors. To be more precise, ν is an upper bound on the fraction of margin errors, and a lower bound of the fraction of support vectors: For regression, the parameter ν replaces ε.

LS-SVM least squares support vector machine algorithm is an improved algorithm of classical SVM. Due to equality type constraints in the formulation, the solution follows from solving a set of linear equations, instead of quadratic programming for classical SVM’s.

Data used in this technique is scaled between 0 and 1. Then it is divided into two parts, part A as training data, and Part B to validate the model obtained after training. A set of pre-processed input and output data is given to the SVR during the training step. After that, the input is given and the SVR calculates the output, which is then compared with the target output to produce an error (e). The model parameters choice (C, ε ,γ ,β and ν) is made with the data A. Best model is re-train with all the data A and validation is done this time with the Database B, the performance of the model are obtained on the basis of statistical indicators nRMSE [8].

Technique 6: Auto-regressive integrated moving average

A time series techniques based on auto-regressive integrated moving average ARIMA is a method first introduced by Box and Jenkins [9] and has now become one of the most popular methods for time series forecasting.

A variation of the classical ARIMA, namely the seasonal ARIMA or SARIMA is used, in order to account for the inherent seasonal effect of the PV power output. The seasonal ARIMA model is generally referred to as SARIMA (p, d, q) x (P, D, Q), where p, d, q and P, D, Q are non-negative integers that refer to the polynomial order of the autoregressive (AR), integrated (I), and moving average (MA) parts of the non-seasonal and seasonal components of the model, respectively.

The SARIMA model is described mathematically as follows:


yt is the forecast variable

θq(B) is the regular AR polynomial of order p

θq(B) is the regular AR polynomial of order q

ΘQ(Bs) is the regular AR polynomial of order P

ΘQ(Bs) is the regular AR polynomial of order Q

The differentiating operator ∇DS and the seasonal differentiating operator ∇DS eliminate the non-seasonal and seasonal non-stationary, respectively. B is the backshift operator, which operates on the observation yt by shifting it one point in time (Bk (yt−k) = yt−k). The term εt follows a white noise process and s defines the seasonal period.

SARIMAX model: SARIMAX model is an extension of the SARIMA model, enhanced with the ability to integrate exogenous (explanatory) variables, in order to increase its forecasting performance. This multivariate version of SARIMA model, called Seasonal ARIMA with exogenous factor (SARIMAX), is generally expressed mathematically as:

image (4)

where image is the vector including the kth explanatory input variables at time t and βk is the coefficient value of thekth ogenous input variable. The stationary and invertibility conditions are equal to those of ARMA models.

Model based on the ARIMA which are SARIMA and SARIMAX. These models have goo error evaluation rate. SARIMA has the rate metrics of nRMSE 18.7% and SARIMAX has the rate of 13.8% [10].

Technique 7: Artificial neural networks

Artificial Neural Networks (ANNs) have been successfully used in various forecasting applications. They are based on the operation of biological neural networks and supposedly possess the ability of a human-like learning process. A typical ANN structure consists of an input layer, a hidden layer and an output layer. Each layer is comprised of neurons that process the input signals and produce an output, while connections between the layers have a weight factor. An ANN easily adjusts to any set of input-output patterns and through a robust training process forms a model function with the minimum possible error.

According to the two models proposed by Vagropoulos et al. [10], models have one output, namely the forecasted PV power generation P(D,H), where D=forecast day, H=forecast hour. Thus, the same formed model is used consecutively for all hours of the forecast day. Many inputs were tested based on the autocorrelation function of the power time series and the strong diurnal periodicity. Two of the ANN models are presented in Table 1.

ANN Model Inputs Output
A P(D-1,H), P(D-1,H-1), P(D-2,H), P(D-3,H), PAVE(H), RMAX(H), R’(D,H) P(D,H)
B ΔP(D-1,H), ΔP(D-1,H-1), ΔP(D-2,H), vP(D-3,H), ΔR’(D,H),ΔR’(D,H-1), ΔR’(D,H+1) ΔP(D,H)

Table 1: ANN models.

Technique 8: Online sequential extreme learning machine with forgetting mechanism

In this technique a proposed short-term PV power online sequential extreme learning machine with forgetting mechanism (FOS-ELM) can constantly replace outdated data with new data. For this technique historical weather data and historical PV data in needed to predict PV power for the next period of time. It has high accuracy with short training period. This model can help the power dispatch department schedule generation plans as well as support spatial and temporal compensation and coordinated power control, which is important for the security and stability as well as the optimal operation of power systems. FOS-ELM is based on extreme learning machine (ELM) because of the computation complexion of ELM is much lower than many other machine learning algorithms. ELM learning speed is much faster than most feed forward network learning algorithms. It has better generalization, and the hidden layers in this algorithm do need to be tuned [11]. So, on the basis of ELM qualities FOS-ELM model can not only improve the accuracy but also reduce the training time. This technique shows that the nRMSE of FOS-ELM is 9.740% [12].

Sensors: Sensors are one of the most important parts in the MPPT. Number of sensors required to implement the MPPTs also affects the decision process. In order to track the maximum power, it is required for the tracker to know the PV inputs (Irradiance and Temperature or I&T) and the PV outputs (Voltage and Current or V&C). Therefore, four sensors are required. However, some MPPTs use modified techniques to reduce the number of sensors. For example, the Open- Circuit method requires only a voltage sensor to track the maximum power (Table 2).

# Tracking Technique Analog/Digital
Sensors Stability Periodic tuning Complexity Speed
1 Fuzzy logic control D V & C Very stable No medium Very Fast
2 Biological swarm chasing D V & C Very stable No High Very Fast
3 PCA and SVM D V & C Very stable No High Very Fast
4 Support vector regression D V & C Very stable No High Very Fast
5 Weighted Gaussian process D V & C Very stable No High Very Fast
6 ARIMA SARIMA D V & C Very stable No Medium Very Fast
SARIMAX D V & C Very stable No Medium Very Fast
7 ANN D V & C Very stable No Low Very Fast
8 FOS-ELM D V & C Very stable No Low Very Fast

Table 2: Intelligence algorithm comparison.


PV power generation is attracting more attention than ever because it is an inexhaustible energy resource but the drawback of PV power systems is their efficiency. So, researchers are trying to get maximum power output through smart and intelligent techniques. Up till now, many techniques have been published, but in this paper only intelligent power point tracking techniques have been analyzed. These are considered advance techniques because of their algorithms. These intelligent algorithms take less time to train the system and are more stable than conventional techniques. With this research success rate, and in the future with development of the PV cell, we might be able to make PV system more efficient, more stable, and more developed than they are today without making PV system high cost capital.


Track Your Manuscript