EM009 » Smart Power-grid Management and Control
Currently, it is not possible to live without electricity. Our lifestyles influence this resource consumption. Indeed, it is obvious that each one of us with his own attitude and mores imposes a certain profile to electrical consumption. This project aims to develop a solution that allows smart remote measurements of electricity consumption. Artificial intelligence (AI) is used to properly analyze the consumption profile targeting individuals identification who certainly have different lifestyles. Moreover, adequate intelligence applied to consumption measurements makes it possible to guess the type of machines used. We also aim to associate with the developed system intelligence, Home automation control planning the residents' actions.
The use of the energy resource is influenced by our lifestyles: using a coffee maker, watching television, working on PC... It is obvious that each individual with his behavior and morals imposes a certain profile on the consumption of electrical energy. Deep and smart analyzes done on this consumption can succeed in present person identification who certainly have different habits. Moreover, each electrical device connected to the electrical grid imposes its own consumption line. Artificial intelligence (AI) applied to the consumption measurements makes it possible to guess the type of machines used. We intend to build real-time and smart telemetry device dedicated to electrical energy consumption. The proposed solution uses an FPGA based board for remote measurements of electricity consumption. Proposed hardware based on the DE10-NANO Terasic board, gauges several electrical signals parameters calculated from instantaneous measurement of the electric power. The interest of the FPGA/SoC device goes toward intelligent analysis for real-time measurements of the electrical current. Instantaneous and real-time electricity consumption measurement associated with correct Artificial Intelligence opens up scenarios to identify installed equipment or even present persons. Examples of use include electrical consumption monitoring hotel rooms, university, institutions…. Measurements can be initiated on-demand remotely or periodically at scheduled times on several parts of the electrical distribution network. The created system can be cloned to be used on a large power grid. Local DE10-Nano based AI analyses can be loaded over the internet or a cloud connection, to improve results and deep learning.
Our proposed solution is done with DE10-Nano as the central module and the heart of the system. The central board is responsible for data acquisition and intelligence. This module also serves as a gateway to the applications hosted on servers or cloud. The used DE10-Nano is placed downstream electrical panelboard for real-time electrical power measurements. Implemented AI and decision making are based first on electrical current measurement analyses.
Besides, our solution brings the prestige of an automated and smart home automation control. Wireless nodes are designed around an ARDUINO platform to assist the DE10-Nano central board to control the connected peripherals of the building. The proposed Arduino based wireless nodes have a derisory cost are easily interfaced because of the DE10-Nano board Arduino Header and Arduino compatibility.
All buildings are equipped with electrical panelboards meeting specific standards and sized according to occupants’ needs. That panelboard relays the electricity to the different circuits of the building. We are using the DE10-nano board first to measure power on both feeders’ outcomes: to supervise the electricity distribution on the diverse parts of the power grid in addition to that coming from the utility provider. The figure-1 below explains Modules disposition and utilization. Figure-2 shows interactions between De10-nano board and environment and implemented functionalities.
Fig-1: Modules disposition and utilization. The DE10-Nano board and connected peripherals are contained by the electrical panelboard. The board measures the electrical power from the feeder and up to 7 electrical circuits of the building. Arduino based connected modules are added to improve users’ wellbeing.
Fig-2: Functional diagram showing environment interactions with the DE10-Nano board. Cyclone V: FPGA is responsible about low-level signals manipulations, HPS implements machine learning, collects data and sends commands to associated electrical devices.
To carry out this project, that involves low-level digital signals processing algorithms, machine learning and communication with external devices. In our hardware and software implementation, we take advantage of both FPGA and HPS components (Cyclone V Soc) as well as the DE10-Nano board architecture.
The Cyclone V FPGA part is involved for DSP tools as digital filters, decimation, spectral analysis, convolution, phased locked loops PLL, phase shift detection, Noise analyses… in addition to real-time and concurrent analysis of instantaneous electrical power measurements.
The Cyclone V FPGA is also used to implement Adaptive-Network-based Fuzzy Inference System ANFIS as an Artificial Intelligence (AI) solution to identify and classify connected equipment to the electrical grid.
The Cyclone V Hard Processor System (HPS) is used through the Linux operating system by means of common and popular tools to exchange data securely with remote servers or cloud (Wi-Fi and Internet). The HPS also implements Machine Learning algorithms exploiting results from the FPGA part for advanced artificial intelligence (AI) analyses. In this case, smart control can be planned based on Machine learning Artificial Intelligence (AI) decisions to improve resident wellbeing and reduce the energy bill.
We have picked the DE10-Nano for its eight channels high-speed and high-resolution analog to digital interface (ADC). Moreover, the board has an Arduino interface offering a simple way for hardware and software integration of multiple electronics shields.
As we have explained earlier, the DE10-nano board has to be placed directly downstream electrical panel board. However, for the proof of concept, As well as for the development and validation of algorithms on FPGA, we are proposing the benchmark shown in figure-3. Our proposed benchmark mimics an electrical power grid within a facility containing two electrical circuits: the first circuit is used for lighting and the second for electrical outlets.
Figure-3: Electrical bench benchmark proposed for the concept proof. The experiment mimics the electrical power grid with two circuits: one for lighting and the other for electrical outlets. Three sensors are used to send continues measures of electrical current to the ADC. We have added a wireless switch and wireless motion detector.
In our workbench, we are using three economical and precise linear hall based current sensors: ACS712-30A to sense the principal feeder line, ACS712-20A to monitor the current on the electrical outlets circuit and ACS712-5A to observe the lighting circuit current (see figure-4 below). The three sensors are interfaced with the DE10-Nano ADC: IN0 to monitor feeder line, IN1 to supervise outlets circuit and IN2 to handle the current on lighting circuit.
Figure-4: Used Sensors detector ACS712 with dissimilar measurement interval to get sensibilities adapted to each electrical circuit applications. With the range +/- 30A we get resolution of 18,5mA, with +/-20A sensor the resolution is equal to 12.21mA and for the last sensor connected with light circuit having range of +/-5A we get 6.6mA as resolution.
The FPGA is responsible for the signal processing of measures representing the consumption of electrical energy. The figure below shows 6 seconds observation of the consumption regarding the lighting circuit.
Figure-5: 6 seconds observation on the lightning circuit having differents Lamps.
The Digital Analog Converter installed in the DE10-nano board is a 12-bit converter. The resolution associated with each connected ACS712 sensor to the ADC is calculated as follows.
Where R is the sensor associated resolution, Vref the reference voltage (always 5V) and S the sensor sensitivity.
The table-1 below summarizes the values for the different sensors ACS712 range we have used:
Table-1: Resolution associated with used ACS712 current sensors
|+/- 30 A||66 mV/A||18,5 mA|
|+/-20 A||100 mV/A||12,21 mA|
|+/- 5 A||185 mV/A||6,6 mA|
The bandwidth of the ACS712 sensors has maximum value of 80Khz hence a minimum sampling frequency of 160KHz is needed. The DE10-Nano ADC operates at a sampling frequency of 500kHz which is largely sufficient. Samples bits are read in serial by the FPGA and it takes 16 cycles of the clock set at 40MHz to read one value. For our software, we are considering to interface the 8 available ADC channels, which reduce the sampling frequency for each ADC line to 312.5KHz, largely sufficient in the context of this achievement.
Regarding the Cyclone V SoC our software solution is distributed between two segments: FPGA and HPS. For the FPGA part we are using Intel Quartus Prime 18.1 Lite free edition IDE. We are taking advantage of IP tools provided as the ADC controller. In the FPGA, regarding our proposed workbench, the software will recover samples from the ADC. Three pipes will be generated simultaneously: Feeder, Outlets, and Lighting. For each pipe similar digital processing are done:
The machine learning on the HPS is implemented based on the TensorFlow framework.