AI based Prognostication of the PV Panel Output Power Under Various Environmental Conditions

Project Title

AI based Prognostication of the PV Panel Output Power Under Various Environmental Conditions

Project Area of Specialization Artificial IntelligenceProject Summary

Electricity plays an important role in the development of any country. Pakistan being a developing country is facing a serious crisis regarding electrical power, photovoltaic (PV) electric power is an excellent option. Moreover for sparsely placed consumers deployment of transmission lines is not suitable economically, again PV is a good option if used with batteries. Another big advantage is PV being a renewable source plays a pivotal role in protecting the environment.

The output power of the PV panel is dependent on the type of solar cells, climatic conditions, and operating conditions. The PV panel performance characteristics and efficiency vary based on various conditions such as irradiance, temperature, shading, dust,  wind speed, PV orientation, maintenance, inverter efficiency, aging, front surface soiling, and breakdown of individual cells. The environmental conditions influence the power output of the PV arrays as well as affect the efficiency of the whole energy conversion system. Hence, predicting the required number of panels and thus estimating the initial cost is a challenge. Therefore, those conditions must be carefully considered before the deployment of any solar PV system to achieve the maximum possible output power and forecasting capital cost required and can also be used in detecting malfunctioning of panels.

This project is based on building up a reliable relationship between the PV system power generation and efficiency, and various environmental factors such as solar irradiance, temperature, dust, and wind, using artificial Intelligence-based machine learning algorithm such as the Artificial Neural Network (ANN) or Support Vector Regressor (SVR) with the feature vectors. Experimental implementation will be conducted to demonstrate the effectiveness of the proposed system.

In the first step, a dataset will be gathered with the help of a data acquisition device that consists of different sensors measuring weather parameters like temperature, humidity, airspeed, air direction, dust, solar intensity, and pressure along with the corresponding power generated. These parameters along with the power generation capacity will be sent to an online database using an Internet of Things (IoT) enabled device. Then, the data will be processed to train a machine-learning algorithm to predict the required number of panels and thus predicting the initial cost.  Moreover, this system will be trained to detect any malfunctioning of the PV panel if the power available and power predicted deviate significantly.

Project Objectives

• To build an internet-enabled data acquisition device that will provide the record for various weather conditions versus power generated
• Development of a user-friendly data platform that can be used by any AI-based algorithm for power generation prediction
• To model a machine-learning algorithm to predict power generation for a PV panel
• To detect malfunction of PV pannel/system

Project Implementation Method Hardware Implementation:

• In the first step, a data acquisition device is made that consists of different sensor like BME280 for Temperature, Humidity, Pressure, BH1750 for light intensity, Devis Anemometer for Wind speed and Wind direction, GP2Y1010AU0F for dust, INA219 for current and Arduino voltage measurement module for voltage, SD Card module as local data storage, ESP8266 for wireless transmission, and Arduino Mega as microcontroller an by using this a three months data-set is gathered
• In the second step, Raspberry Pi will be used to predict power in real-time after the dataset has modeled  a machine learning algorithm 

Software Implementation:

• MATLAB for Data Pre-processing such as interpolating missing sensors data and Feature engineering in order to transform the data into useful information
• Rapid Minor for modeling a machine learning algorithm
• MATLAB Simulink for simulation

Benefits of the Project

• The exact amount of power generation from a PV system can not be estimated due to frequent changes in weather conditions. By predicting the power generated we can ensure the best utilization of resources and helps in determining the exact initial cost required for a defined power
• It helps in detecting malfunctioning of PV panels by comparing the power generated against power predicted
• Can help in reduced energy generation from furnace oil and gas plants to combat climate change and its impacts

Technical Details of Final Deliverable   Block Diagram:

The architecture of our project can be described from the following figure.

It can be divided into the following parts:

Sensors Module

Sensor module is a part of the data acquisition device that contains the following sensor:

1. BME280 for Temperature, Humidity, Pressure measurement using the I2C interface with Arduino Mega 2560.
2. Davis Anemometer for wind speed and direction measurement. Wind speed is read on analog read pin of Arduino where wind direction is read digitally on digital pin from 0-360 degrees.
3. Optical Dust Sensor - GP2Y1010AU0F for measuring the level of dust whose value is read on the analog pin of Arduino
4. BH1750 for measuring the intensity of light from 0-64560 lux. But as to measure the full range of intensities we introduced a certain amount of darkness while that value is read using the I2C digital interface.    
5. RTC3231 is a real-time clock that is used to keep the count of time every time a dataset is generated

Photovoltaic Panel:

It is a combination of four 30 Watt MAXPOWER monocrystalline PV Panel whose electrical characteristics are as follow:

Maximum Output Voltage	18.29 V	Open Circuit Voltage	22.07 V
Maximum Output Current	8.21   A	Short Circuit Current	9.05   A

Voltage & Current Measurement:

The output current of the PV Panel is measured using INA219, a sensor with a digital interface with Arduino.

Where Output Voltage of the PV panel is measured using the Arduino voltage sensor whose value is read on the Arduino analog read pin.

As there is no load measurement case, we need to measure both separately. Thus that is performed using a two-channel relay module.

Local Data Storage:

Every time a data-set is generated, it is saved in SD Card that is connected to the SD Card module whose connection with Arduino is an SPI(Serial Peripheral Interface).

Wifi Module for Online Transmission:

ESP8266 is used to transmit data online to display on the web-page.

The whole of the data acquisition can be described from the following scenario diagram.

Data Engineering & Modeling:

Once a dataset is collected it is engineered to convert it into useful information:

• Interpolating missing sensor data using MATLAB
• Finding mean, standard deviation & variance, etc. of data

When the dataset is ready it is used to model the machine learning algorithm like Recurrent Neural Network and  Support Vector Regressor (SVR) with the feature vectors and model that performs best based on the parameters

like Mean absolute error (MAE), Mean bias error (MBE), Root mean squared error (RMSE), Normalized root mean squared error (nRMSE), Correlation coefficient (????).

Real-Time Power prediction & Capital cost estimation:

Once the model has been trained then this is used to predict power output from photovoltaic modules in real-time. It takes the input from data acquisition real-time & then depending upon the model it forecast power.

Maximum Output Voltage

Maximum Output Current

Final Deliverable of the Project HW/SW integrated systemCore Industry Energy Other Industries Education Core Technology Artificial Intelligence(AI)Other Technologies Internet of Things (IoT)Sustainable Development Goals Affordable and Clean Energy, Industry, Innovation and Infrastructure, Responsible Consumption and Production, Climate ActionRequired Resources

Elapsed time in (days or weeks or month or quarter) since start of the project	Milestone	Deliverable
Month 1	Literature Review	architecture diagram
Month 2	Study of different controllers to perform the task	Selection of Controller
Month 3	Study of sensors required	Selection of Sensor
Month 4	Interface of sensors to controller	Data acquisition device
Month 5	Study of machine learning algorithm	Selection of suitable machine learning algorithm
Month 6	Data-set collection & mini computer raspberry pi introduction & Data engineering	sorting data as per requirement of algorithm (RNN, SVR, etc)
Month 7	Modeling machine learning algorithm	power prediction
Month 8	Final hardware implementation on raspberry pi	Real time power Prediction
Month 9	Final thesis and documentation	Report
Month 10	Final thesis and documentation	Report & Possible publications