Solar panels simulation data generated using LTSpice under different operating conditions

Abstract

This paper presents a detailed description of the circuit simulation data obtained using LTspice, a high-performance SPICE simulation software for easing the simulation of solar panel circuit data. The data represents the photovoltaic modules with different configurations and cells placed in series and parallel. The data are obtained from automated Python creation and simulation of PV cells in specified formats. The collected data are then organized in a CSV (Comma Separated Value) file. Each file contains properties of the photovoltaic cell, such as individual cell voltage, current, and others. The data are then used to evaluate each of the photovoltaic modules of an array.

Specifications Table

Subject	Computer Science, Engineering
Specific subject area	Artificial intelligence, Photovoltaic energy, Optimization, Machine learning
Type of data	Table Image Figure
How the data were acquired	Data collection is performed by running SPICE-based equivalent PV module simulation and creating multiple CSV files with a total of 6965,234 data points. For simulating the PV cells using SPICE, we considered the area of PV cells to be 49 cm2 which gives out an open circuit voltage of 1.025 V, and a short circuit current equal to 30.5 mA/cm2. The saturation current density of the two diodes, D1 and D2, shown in Fig. 1, is equal to 0.16 aA/cm2 and 1.2 pA/cm2, respectively. The RS and RP values for the equivalent model shown in Fig. 1 are 28mΩ and 100 KΩ, respectively.
Data format	Raw
Description of data collection	This article presents an extensive dataset that is simulated by using the tools in Python and SPICE tools such as LTSpice. The various partial shading, open and short circuit conditions were emulated by utilizing SPICE netlist which is generated by using Python. Later by using the LTSpice simulation (Version 17.0.35), various datasets are generated for different configurations at different operating temperatures.
Data source location	Institution: California State University City/Town/Region: Fullerton Country: United States Latitude and longitude (and GPS coordinates) for collected samples/data: 33.8823° N, 117.8851° W
Data accessibility	Primary Data With the article Repository name: Mendeley Data Data identification number (DOI): 10.21227/fjbq-0321 Direct URL to data: https://data.mendeley.com/datasets/3fr92f4xy9/1

Value of the Data

• •The data presented in this work can interest a broader community of researchers involved in designing and modeling machine learning algorithms for optimizing solar panels.
• •With this data, the readers can evaluate the behavior of PV systems under different shading and fault conditions at various temperatures to better understand the performance of PV panels in different conditions.
• •In this research, we compare a variety of machine learning (ML) algorithms to identify faults in PV systems. These results can be helpful for comparing with other ML algorithms to enhance the performance of PV based systems to power drones, lookout towers, small electronic devices, embedded systems, and many others. Drones have shown their effectiveness during natural or human-made hazards such as forest fires, nuclear accidents, floods, and others. In such operation before the deployment of solar powered drones, the presented data will enable to test the performance of PV modules in different hazardous operating conditions. Even smart sensors installed at a remote location can be powered by PV based system. The presented dataset is useful for simulating its performance in various operating conditions.
• •Recently, machine learning [1], [2], [3], [4], [5] and fuzzy logic-based [6], [7], [8], [9] maximum power point tracking (MPPT) have been used to ensure the PV systems can operate at optimal efficiency. All the techniques require a dataset to test their algorithm's effectiveness before deploying it on an existing PV system. The data presented in this work is beneficial in testing the MPPT algorithm in a realistic virtual environment.

1. Data Description

The collected dataset presents ten attributes that describe the solar panel. Each solar cell is represented as an individual instance in the training dataset. The solar cells are modeled using the doubled diode based equivalent model shown in Fig. 1. For every instance, a unique serial number is assigned, incremented by one, and includes features associated with a PV cell.

Fig. 1.

Equivalent PV cell model.

Fig. 2 presents a snippet of the attributes of the solar panel that are used as descriptive features and fed as input to the machine learning algorithms. The complete dataset used in this work is available at the Mendeley Data repository [10]. The descriptive features are mapped using a hypothesis and the relationship between the input features and the target label is captured by the mapping function. The mapping function can be used to make predictions on new unseen data. Fig. 3 highlights the correlation between the input features. The correlation matrix highlights that the ‘series cells‘ feature is highly correlated with voltage and parallel cells.

Fig. 2.

Input features for machine learning models.

Fig. 3.

Correlation matrix for the input features.

Additionally, power and current are highly correlated as well. Fig. 4 shows the relationship between variables categorized by their fault types. This work explores three types of faults in solar panels: shaded, short circuit, and open circuit. In the dataset, we use 0 for shaded, 1 for short circuit, and 2 for open circuit. Fig. 5 presents the features plotted based on temperatures ranging from 20 to 50 °Celsius.

Fig. 4.

Feature Visualization plot by type.

Fig. 5.

Feature visualization plot by temperature.

2. Experimental Design, Materials and Methods

For the ML based classifier to be effective, it is essential to have an accurate dataset with plenty of data points to train the ML classifier. However, generating such an extensive dataset with experimental data with all possible operating conditions probable during the mission is tedious. Hence, to simplify this lengthy task SPICE modeling technique can be used to generate the dataset. In [11], to generate the dataset for different operating conditions of PV module, Python, LTSPICE XVII (Version 17.0.35) [12], and 2-diode based equivalent PV module modeling technique is used, as shown in Fig. 1.

To create a dataset for training the ML algorithm, the following conditions include solar irradiation, various PV panel configurations, fault conditions (Short, Open), partial shading conditions, and temperature. The different PV configurations used resulted in the simulation of 6965,234 different scenarios. For emulating the partial shading conditions, we consider the solar irradiation from 500 to 1000 W/m². An ideal unshaded PV cell is with solar irradiation of 1000 W/m². For emulating the distinct types of shading conditions throughout the day, the solar irradiation of shaded PV cells varies from 900 W/m² to 500 W/m². Later, the Python program designed for this paper programmatically creates all the required scenarios via the following equations:

$${\mathrm{C}}_{\text{open}[\text{Ns},\text{Np}]}=\left({\mathrm{N}}_{\mathrm{P}}-\mathrm{i}\right)\mathrm{x}{\mathrm{N}}_{\mathrm{S}}$$ (1)

$${\mathrm{C}}_{\text{short}[\text{Ns},\text{Np}]}={\mathrm{N}}_{\mathrm{P}}*\left({\mathrm{N}}_{\mathrm{S}}-\mathrm{j}\right)$$ (2)

$${\mathrm{C}}_{\text{shade}[\text{Ns},\text{Np}]}={\mathrm{N}}_{\mathrm{P}}*{\mathrm{N}}_{\mathrm{S}}$$ (3)

$${\mathrm{S}}_{\text{Ns},\text{Np}}={\mathrm{C}}_{\text{open}[\text{Ns},\text{Np}]}+{\mathrm{C}}_{\text{short}[\text{Ns},\text{Np}]}+{\mathrm{C}}_{\text{shade}[\text{Ns},\text{Np}]}$$ (4)

$${\mathrm{S}}_{\text{tot}}=\Sigma \left({\mathrm{S}}_{\text{Ns},\text{Np}}\right)*{\mathrm{T}}_{\mathrm{s}}*{\mathrm{I}}_{\mathrm{s}}$$ (5)

where C is the number of different configurations for N_P is the total number of PV cells from parallel and N_S is the total number of PV cells in series, which are considered from 1 to 10. The variable i in the Eq. (1) is given by 0 < i ≤ N_P −1 and variable j in Eq. (2) is given by 0 < j ≤ N_S −1. For the short and open circuit emulation, a number of the PV cells in series and parallel were electrically removed, respectively.

Eqs. (1), (2), and (3) describe how the number of necessary simulations is determined for open, short, and shade conditions, respectively. Eq. (4) is the combination of (1), (2), and (3) and is the total simulations to be run based on the various PV panel configuration, with (5) being the total amount of simulations to run, where T_s is the number of temperature scenarios and I_s is the amount of solar irradiation intensity scenarios. Once per PV configuration is determined from (4) for a set [N_S,N_P], the Python program generates SPICE netlist (.cir) files for simulation. After all PV panel per cell configurations have been created, LTSPICEXVII is run via the Python program across all .cir files, producing .raw files. The files contain output voltage, generated current, and total power per cell, which are collected via the Pandas library into a single .csv file, which is used in ML processing. The subsequent steps involve using ML tools in preprocessing, training, testing, and classification. The use of, which are collected via the Pandas library into a single .csv file, which is used in ML processing. All steps involved in generating the dataset are shown in Fig. 6.

Fig. 6.

Steps involved in Dataset Generation.

Ethics Statements

These data are primary data and do not include human subjects, animal experiments, or social media platforms.

CRediT Author Statement

Kanika Sood: Machine Learning Modelling, Writing, Methodology, Software, Conceptualization; Rakeshkumar Mahto: Investigation, Validation, Conceptualization, Data curation, Writing; Nathaniel Ruppert: Visualization, Data analysis, Data Simulation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data Availability

• Simulation Dataset of Partial Shading and Fault of a Photovoltaic Module (Original data) (Mendeley Data).