Integrating Microsoft IoT , machine learning in a large scale power metre reading

Due to fast technological progress in the power engineering field, the need for new information/communication technologies is more and more underlined. e-Learning has become a viable alternative to traditional teaching/learning techniques, adopted especially because of the advantages offered by the possibility of continuous training. This paper presents a Microsoft Internet of Things Platform for a very large scale smart power metre reading, used for training operative staff of the Distribution Network Operator, but also to help end-users to control their electrical energy consumption. The strength of this platform for the Distribution Network Operator is that the read data can be used for energy forecast, which is very useful for the future energy consumption optimisation. The platform can be reached via the Internet using a user name and password. A comparison between the results provided by classical teaching/learning methods and the ones achieved using this platform is presented.


Introduction
In the last period, many improvements were done in the metering area.The new digital metres are capable of storing much more information concerning how the end user is consuming the energy and are capable to automatically or on request transfer this information to the energy distrusters.These new digital metres are called smart metres because they are offering the possibility to the energy distributor: to better understand the needs of the customers, to better invoice the exact consumption and to better forecast the future consumption.
In the same time, in order to be able to use the information possible to obtain from these smart metres, new platforms were developed in order to be able to connect to smart metres, to store the historical data and to manipulate the received information.One of these new platforms that provide the capability to connect, store and manipulate the information received from smart metres is Microsoft Internet of Things (IoT) platform.
Our paper presents a training system that uses the Microsoft IoT platform and a set of simulated hourly energy consumption data set (similar with the ones possible to obtain from smart metres) that predicts the energy consumption for the next day in order to train students or distribution networks operators on how to use such a system.Because a lot of effort and investment are put in the next years to utilise as many end users as possible with smart metres, as future utilisation of the actual system with minimal effort, the simulated data set can be replaced with the information received from real smart metres and the prediction offered by the system can help the distribution networks operators to better manage and plan the actions needed to be done in order to operate their networks in an optimal way (Jigoria-Oprea, Simo & Borlea, 2014).

Integration in real environment
The actual situation and future plans to have a smart metre for each end user are using a very hybrid solution in terms of types of equipment.There are smart metres that are accessible over IP, GSM or radio.The actual situation and planned one propose that concentrators are used to communicate with smart metres and provide the received information to the next systems that will store and manipulate the data sets.Most of the concentrators are compatible with industry standards but still some of the equipment will have native integration with Microsoft IoT platform and some will don't.For the real environment, our proposed solutions take into account this multitude of equipment that will exist in the field and propose a solution that will be able to handle the situation.In Figure 1, the main components of the proposed solution and how they interact is presented (Metering, 2015).

Integration in training environment
In the training environment, the connectivity with the real smart metres and concentrators was replaced with a computer program that simulates the measurements from 100,000 smart metres.In Figure 2, the main components of the training solution are presented.

Figure 2. System main components and interaction in training environment
The program that simulates the measurements from 100,000 smart metres is generating the hourly consumption based on the following changeable inputs: • Types of equipment used and their consumption dependence on the outside temperature.
• Percentages of consumer types from the total number of estimated consumers.We use eight consumer types in our simulations.
And based on the unchangeable inputs: • Historical hourly temperature measurements obtained from specialised websites for the last 365 days.
The simulated data set is consumed by the raspberry PI equipment and transferred to the Azure IoT Hub, stored in an HDInsight Data Lake Service and analysed by a predictive analytic solution implemented in a Machine Learning Azure service that based on the data set received from raspberry PI device, and the temperature prognoses for the next day is providing an estimated hourly energy consumption.Finally, the results are presented to the trainee on a web interface (Historical temperature measurements, 2016).Some examples of the parameter used by the simulation program are presented in Figure 3 (Next day prediction, 2016).

Advantages for using the training system
The following advantages are to be mentioned:

Case study
The assumption on which our case study is based is that plausible results can be obtained by applying modern teaching/learning techniques for operative staff training, regarding the use of the developed platform and its features.Also, this case study shows that the results are better if modern and traditional methods are combined.
The participants in this study are the operative staff of the Distribution Network Operator.There were 60 participants: 24% women and 76% men.Their average age was 36 years old.The participants were divided into three groups of 20 people, heterogeneous in terms of their levels of knowledge.
The study lasted 6 weeks, during which 60 participants were divided into three groups of 20 (OS1, OS2 and OS3).Individuals from group OS1 attended a training session, where the information content was submitted by the traditional mosaic method.The mosaic method is used for the practice of individual and collaborative learning.It is used by experts as an aid for teaching and requires a flexible algorithmic process within small activity groups (Rosenberg, 2011).The whole group OS1 was divided into small groups of four, each time in a different order to change their roles and tasks (Simo, Barbulescu & Kilyeni, 2014).
The steps in this process are the following: a) the formation of heterogeneous groups; b) the presentation of the topic and its division into sub-themes; c) the formation of expert groups, presentation of instructions and clarification of tasks; d) the study of material in the experts' groups and preparation of responses to those in the home or original group (this is the time for experts to teach their teammates); e) the restoration of the original groups and the presentation of answers to the teams of experts; f) individual testing; g) assessment of teamwork and recognition of their merits; h) summarise the position and arguments of a workgroup and the writing of individual work.
Group OS2 received the material electronically, tutorials and visual demonstrations.During this period of study, they had never seen any trainer.The participants studied by themselves through the informational content and solved a number of the proposed problems.This kind of approach is also the case of end-users, which after the platform launch on the Internet will not have access to a trainer, only to tutorials.
Group OS3 also received the material in an electronic format, the same content as participants in group OS2.However, after independent study, the forum platform was available for online questions addressed to trainers or other participants.
At the end of this training period, a final evaluation was done to measure the level reached by each group and each participant, and the results obtained were compared by different teaching-learning techniques.The grades of participants were divided into three categories: small grades representing the grades in the [1.. 4] range, medium grades representing the grades in the [5.. 7] range and high grades representing the grades in the [8.. 10] range.

Results and discussions
The obtained results are synthesised in the following Table 1.For a better understanding, the grades distribution is presented in Figure 4.As the next step were analysed differences between groups, regarding the grades distribution.
Furthermore, these obtained values were analysed using the chi-square test.The expected results are presented in Table 2 and Figure 5.The p-value for the calculated 2 is 0.1756, so our hypothesis is confirmed.The deviation is small enough that chance alone accounts for it.This value means that there is a 17% probability that any deviation from expected results is due to chance only.Of course, this is within the range of acceptable deviation (Edu, 2016).Noticeable differences can be observed between groups OS1 and OS3.Participants from group OS3 obtained fewer low grades and more medium grades compared to group OS1; however, there is no difference in the highest grades.Differences between groups OS1 and OS2, and groups OS2 and OS3 were statistically insignificant.

Conclusion
The results of the case study lead us to conclude that a combination of traditional and modern methods is more successful for operative staff training.For example, participants in group OS3 were trained online by modern methods but were offered daily support by the presence of a trainer via our online platform.
We believe that by using traditional methods combined with modern ones, participants can achieve better performance.

Figure 3 .
Figure 3. Parameters used by the simulation program

Figure 4 .
Figure 4. Graphical representation of the group's results Figure 5. Chi-square test results