2.1 Problem Statement

We can summarize our approach as follows. In companies that generate, distribute, or consume electrical energy, there is a large volume of data resulting from their operations that can generate relevant information. Currently, such data is not being used to improve their operational and commercial management.

Source: Zhang, 2018

Figure 1 Data sources in a power grid. 

Source: Adapted from Alavikia and Shabro, 2022.

Figure 1 IoT applications to support Smart Grids. 

For example, to support our approach, ref. [⁶] showed different data sources in an electrical network that indicate the existence of a large amount of data that can be used for decision making.

Fig. 1 shows that there exists data from the monitoring of electrical load and energy quality within the group of structured and unstructured information. These data are relevant for adequate management of the electricity supply, both for generators and distributors as well as for consumers of electricity. This is because it has an impact on the economic aspect and on the adequate use of the equipment.

Likewise, in [⁷], it was stated that the latest generation power systems require the implementation of smart grids (SG) at all levels (generation, distribution, and consumption). These networks have high device connectivity owing to the increasingly accepted use of the Internet of Things (IoT). Such high availability of connections and the permanent exchange of data generates a large volume of data that requires techniques to help conduct adequate analysis for making timely decisions [⁸].

In [⁹], it was shown that from the point of view of telecommunications, all the data of an electrical system are derived from three communication networks and "can be mapped into Home Area Network (HAN), Neighborhood Area Network (NAN), and Wide Area Network (WAN)" [⁹], as shown in Fig. 2. The HAN includes wireless devices, electronic meters, gateways of homes or medium and small industries. The NAN includes the parts of the distribution networks: disconnectors and totalizing meters. The WAN includes gateways between networks of generator control centers, main transmitters, and electronic meters of large industries.

The development of smart grids in Peru has been linked to improvements in energy efficiency. This is evident from the improvements in terms of energy efficiency and from legal initiatives that have been proposed. For instance, in 2010, initiative 4335 proposed the promotion of smart electricity grids as a method to increase efficiency in the electricity system. This was intended to optimize the monitoring and control of electricity consumption, promote greater use of renewable energies and distributed generation, and grant users' greater benefits for lower or equal cost [¹⁰].

There is an interactive energy efficiency system (SiEE) in Peru, where it has been reported that since 2017, public electricity distribution companies (EDES) must develop investment plans in innovative technologies that improve the efficiency of distribution networks and implement pilot plans for the installation of smart meters. Additionally, transition proposals from traditional grid architectures to smart grids have been shown [¹¹].

In [¹²], the authors stated that the challenge for Peru is to propose the inclusion of smart grids as an important objective for the competent authorities of the electricity sector, change the regulatory framework, promote research, and include consumers in the development of knowledge regarding the benefits of smart grids. In this paper, we adhere to these ideas.

The inclusion of IoT in the industry, has been explained in [¹³] and mentioned in [¹⁴]. It involves interaction or communication among devices that generate massive data, thus requiring control and management strategies to derive greater benefit from them. Such strategies can be implemented and a large amount of data can be processed and used quickly and efficiently.

The aforementioned statements support our assertion of the existence of a large amount of data (big data) at all levels of the electricity business. Next, we present some solutions related to data analysis in companies in the electricity sector that highlight the importance of these applications.

In [¹⁵], the relevance of data analysis and machine learning for non-intrusive load monitoring (NILM) has been discussed, allowing adequate control of the electrical energy consumption of critical equipment in residential, commercial, and industrial facilities.

In [¹⁶], the use of data analysis for the management of a photovoltaic system was shown. Using advanced data analysis applications (machine learning and statistical techniques), the authors have obtained important conclusions.

In [¹⁷], experts from all relevant organizations and institutions, including academia and industry, met to discuss big data analysis applications for energy systems. They concurred that becoming a data-based electrical company is inevitable in the era of the internet, cloud storage, smartphones, and distributed resources. Furthermore, the analysis of advanced data involves ongoing innovation, which is plausible by unlocking information that has not yet been processed, while advanced big data analysis results in the conclusions of estimation, prediction, diagnosis, and forecasting of historical and real-time data flows. This can provide more refined information on planning network operations.

Along the same lines as stated in the previous paragraph, energy monitoring systems have achieved savings in energy billing between 15% and 35% according to feedback (visualization is indirect or direct), as shown in [¹⁸]. In energy monitoring systems, energy consumption is visualized and data analysis algorithms are used to examine the behavior of different devices in a residence,

Hence, it is evident that in electricity generation, transmission, and among distribution companies and electricity users (residential, commercial, and industrial), there is a large amount of data with the potential to generate valuable information for decision-making. Therefore, developing solutions in this area of research is a current trend that can also generate significant savings in maintenance and operation management for all companies in the sector, from generation to end users.

Table 1 Subject, object, problem and research question of the present study. 

Source: The authors.

BI is a system that allows users the possibility of converting data into information. As defined by the Data-Warehousing Institute, business intelligence or BI includes "tools, technologies, and processes required to turn data into information and information into knowledge and plans that optimize business actions" [¹⁹].

The extraction of valuable information and discovery of underlying data patterns allows BI to make fast and effective decisions for different organizations [²⁰]. Therefore, companies that manage electricity should not be an exception.

A very interesting issue is presented in [²¹], stating that Business Intelligence and Analytics (BI&A) is mainly a subfield of decision support systems (DSS), which aims at converting more data into a thorough vision but by severing its link with the founding literature by DSS.

DSS are information systems that aim to improve decision-making based on data analysis [¹⁸]. The interesting aspect of such an approach is the fact that, by regarding BI&A as a subfield of DSS, gaps and research opportunities in the field of BI&A arise, which can utilize the foundational concepts of DSS.

For this reason, in this study, we propose a methodology to develop a BI solution in a Peruvian electricity generation, distribution, or consumption company that can help improve its operational and commercial management by considering the regulations in force in 2020.

In Table 2, we show the summary of our approach.

Table 2 Summary of the analyzed methodologies 

Source: The authors.

2.2 Existing methodologies - background

In this section, we present four methodological antecedents available for the final deployment of a solution that fits our requirements.

The first methodology, generally accepted in the industry as one of the best practices, is the Kimball life cycle raised in [²²], which follows a general doctrine to focus on business value, collaborate with the company and develop incrementally.

The roadmap of this methodology is shown in Fig. 3. The first step is the definition of business requirements, followed by parallel developments. On one hand, the design, development and deployment of the BI application must be supported in an adequate dimensional modeling of the data. On the other hand, a physical design as well as design and development of an ETL solution (data extraction, transformation, and loading) is required. Kimball's approach is very robust, but in our opinion, it can be simplified depending on the characteristics of the solution we are looking for, even more so when it is applied to a small or medium-sized company.

Source: Kimball, 2013.

Figure 2 Kimball life cycle diagram. 

Source: Adapted from Ramos, 2021.

Figure 3 Steps to create a BI solution. 

Another methodological alternative is the one proposed by Ramos in [²³]. The steps outlined in Fig. 4 are logical and are in accordance with our objectives. These coincide with the first methodology regarding defining the business questions and as they are exclusive steps for the deployment of BI solutions, there are no parallel processes. We must specify that in the "modeling" step, the method refers to the modeling of the data.

Source: Huber, 2019

Figure 4 Phases of the CRISP-DM reference model. 

Another methodology introduces two additional closely related concepts: business analytics (BA) and data mining (DM).

The subtle difference between BI and BA is stated in [¹⁹]. Both terms are used interchangeably because they provide similar challenges in terms of implementation and adoption. In addition, the same article regards business intelligence as an umbrella field covered by business analytics, such that business analytics results in an important aspect of business intelligence, as stated in [²⁴].

Also, in ref. [²⁴], data mining (DM) is regarded as the process of "knowledge discovery", which exists in the data base and can be of used in decision making. Thus, data mining provides insight into past and present scenarios and helps comprehend possible future results that can supply effective outcomes and assist in discovering unknown facts present in the database.

In 1996, for the development of a DM solution, four leaders of the emerging DM market conceived the "CRISP-DM (Cross-Industry Standard Process for Data Mining)" [²⁷] model, as a generic, documented and free access model that allowed organizations to achieve better data mining results and promote best practices in the industry [²⁵,²⁶].

Fig. 5 shows the phases of a DM process, proposed by CRISP-DM, in which the six phases are as follows: "understanding the business, understanding the data, data preparation, modeling, evaluation, and deployment" [²⁷].

Source: The authors.

Figure 5 Raw data - Screenshot of electrical parameters for seven days. 

The authors in Ref. [²⁶] have provided an explanation for Fig. 5, in which the sequence of steps and the interrelation between them can be observed. Additionally, the cyclical nature of the process is represented by the large ellipse. This indicates that after some solutions have been identified based on the knowledge found, new solutions can be generated from the previous discoveries.

The objective of our study is to find a method to implement a business intelligence framework that can be applied to electricity companies or electricity users, especially for small or medium-sized companies. Using this criterion, we have evaluated the three methodologies, as shown in Table 3.

2.3 Methodology used

Considering the CRISP-DM method as a reference, we have included an additional step prior to preparing the data. This is because it is useful for the target companies to be informed about this type of analysis and to define the data model prior to finding the process model to be applied. The following is a methodology based on [²⁶].

2.4 Application of the methodology.

To apply this methodology to a specific case study, we have adopted the following steps:

Raw data were obtained from the production files of the generating company for one plant and the data were reviewed and pre-processed to select an adequate model. In this case, the methodology was applied to specific objectives, but the objective was to validate whether the steps could be followed.

• The methodology was applied to the company's available data, using Power BI for the data modeling, extraction, transformation, and loading phases.

For the transformations, the M language was used (transparent when implemented from the application interface) and new measures or columns were generated in the DAX language.

3.1 Step One: understanding the business

The company is a seed-processing plant that works 24 h a day. To manage its electricity supply, it operates with its own generator set during peak hours. The following objectives were set for this study:

3.2 Step Two: understanding the data

The data were obtained from a network analyzer that reported electrical parameters such as voltage, current, power, flickers, and harmonics. These parameters were measured every 15 minutes for 1 week. The data was in the source format. See Figure 5.

3.3 Step three: defining the type of analysis

According to the characteristics of the business and the objectives sought, two types of analysis were carried out: descriptive analysis and prescriptive analysis.

The description showed us the behavior of the main electrical parameters involved in the evaluation of the quality of the electrical supply.

The prescriptive analysis provided recommendations to improve behaviors that affected the quality of energy and raise the cost of energy billing.

Source: The authors.

Figure 6 Screenshot of data transformed after cleaning, ordering, transformations, and calculations. 

Source: The authors.

Figure 7 Screenshot of data model does not require relationship between tables - A measurement table was generated. 

Source: The authors.

Figure 9 Screenshot of the visualization of tension behavior. 

Source: The authors.

Figure 10 Visualization of the behavior of the power factor and the instantaneous power. 

3.4 Step Four: prepare the data

Once the types of analyses to be applied were defined, the data required transformations (elimination of columns, creation of calculated columns, and combination of columns). A single-table data model was loaded, which was ultimately a time series. See Figure 6.

Modeling of the data in this case was not very complicated because it included a table with the measured parameters and tables of calculated measurements. This can be observed in Fig. 8.

3.5 Step Five: modeling the data

Descriptive analysis was carried out using diagrams such as those shown in Figs. 9 and 10.

Fig. 9 shows several instants of time with a voltage value above the maximum permissible limit by standard (+5% of the nominal voltage) plotted with the upper horizontal line. Likewise, the percentage of time intervals with voltage out of range is almost 40%, which is quite high.

Fig. 10 shows that the power factor in the whole measurement is below the minimum limit that avoids payment of compensation for reactive energy consumption.

Hence, descriptive analysis allowed the diagnosis of problems of poor voltage quality and inadequate reactive power consumption.

In these cases, the prescription is to verify positions of the transformer TAPs and calculate a bank of capacitors to compensate for the high consumption of reactive energy. A more in-depth prescriptive analysis, such as that used in [²⁹], proposes a selection strategy for the steps in a capacitor bank. Fig. 11 shows the results of the description of the behavior of the power factor and, in the lower graph, the result of the application of a clustering model of the power factor with six clusters.

3.6 Step Six: evaluation

The visualizations generated using the processed data model for descriptive purposes allowed us to answer the initial questions. The quality of the electricity supply was unacceptable. It is shown that the tension has many values above the maximum limit of 5% required by the Peruvian quality standard. In addition to being out of range, these values may cause damage to electrical equipment. Therefore, immediate correction is required.

Likewise, it is shown that the power factor is below the economic operating limit (0.96), indicating that during the entire period analyzed, the reactive energy consumption has exceeded the limits, and therefore, they are being billed.

By applying a clustering model, we can prescribe a 6-step capacitor bank, which is more effective than a conventional one-step fixed bank. Observing Table 3 with the results, we can define a bank with the theoretical steps shown in Table 4.

Source: The authors.

Figure 11 Screenshot of the clustering of active and reactive power consumption according to the Power Factor to 6 Clusters. 

Table 3 Compensation capacitive reactive powers calculations 

Source: The authors.

Table 4 Compensation steps of the capacitor bank (theoretical). 

Source: The authors.

3.7 Step Seven: implementation

The implementation in this case involves the presentation of these results in a report that shows the findings and implementation of the prescriptions made: movement of the TAPs of the transformer to adjust the voltage, as well as the installation of a bank of capacitors in six steps.

These steps must be adjusted to commercial steps. Through calculation using averages and maximum powers where a 120 kVAr capacitor bank is prescribed, the capacitor is inferred to be oversized and can cause overvoltage when operated in a single step.