Data Driven Techniques for Energy Saving in Manufacturing Companies
As a researcher in the Industrial Assessment Center (a lab funded by the US Department of Energy), energy analysis for efficiency and waste reduction has been identified as driving factors in the reduction of Co2 emissions. More importantly, these factors are behind the adoption of an eco-friendly path to mitigate the ills of climate change. Sustainable energy solutions can only be birth by energy serving and adhering to policies that make for a reduction in power consumption of equipment in manufacturing companies and facilities.
With the rapid development of sensor technology, wireless transmission technology, network communication technology, cloud computing, and smart mobile devices, large amounts of data has been accumulated in almost every aspects of our lives. Moreover, the volume of data is growing rapidly with increasingly complex structures and forms. Continuous improvement in energy efficiency plays a critical role in reducing environmental impacts. Driven by regulations and cost pressure, more and more manufacturing companies have implemented energy conservation technologies and systems to improve their energy efficiency as an important part for continuous improvement of their sustainable performance. Other than the compulsory actions by legislation like the ECA, many companies set internal target and implement voluntary programs to improve energy efficiency for reduction of cost and environmental impact. Both the compulsory and voluntary programs require continuous improvement. To meet each year’s target, energy managers and their team have to identify the areas that have both high energy saving potentials and economically viable ways to implement.
In the energy sector, large amounts of energy production and consumption data are being generated and the energy systems are being digitized, with the increasing penetration of emerging information technologies. The innovations brought by big data are changing the landscape of traditional industry. Currently, manufacturing companies are facing various challenges such as challenges in operational efficiency and cost control system stability and reliability, renewable energy management, energy efficiency and environmental issues, as well as consumer engagement and service improvement. To better deal with these challenges, energy big data analytics provide new opportunities by achieving smart energy management. Specifically, to achieve the goals of efficient power transmission within machines, dynamic power distribution and rational electricity consumption,energy polices that incorporates distributed generation resources and innovative storage solutions have been proposed.
Data Analytics Techniques for Energy Savings
The data analytics approach is carried to dig out energy inefficiency occurrences, analyze the occurrences for best achieved energy efficiency frontiers, and quantify their improvement potentials.
Energy Improvement Frontiers
For energy efficiency improvement programs, targets to meet energy demand are set for individual organizational units. Each organization unit has its baseline and potentials due to functional differences. Optimal energy targets are set by energy consultants ensuring peak demand is reduced and energy bills are significantly reduced. To identify the energy efficiency frontiers, the CCR (Charnes, Cooper and Rhode) model of the DEA (DataEnvelope Analysis) methodology can be applied.
Correlation Analysis
The time-series matrices contain rich information and knowledge of the operations of individual machines, a group of machines, or the entire factory. The value of the parameters are obtained independently through meters/sensors. In order to understand their correlations and dependency, relevant statistical method is needed. Pearson correlation coefficient (PCC), denoted as r, provides a measure of linear correlation between two quantitative variables, such as time-series datasets, X={x1, x2, …xn} and Y={y1, y2, …, yn}
Quantification of Energy Saving Potentials
As the efficiency frontier represents the achieved best efficiency, the maximum energy saving potential can be taken as the difference between the current energy consumption and that if all the output occurrences, , of machine u become as efficient as the efficiency frontier, i.e., Esaving=where is the time interval of data reading (t=1, 2, …, n), is the energy efficiency frontier function from regression analysis, is the output parameter at time, and T is the time period. Iterating the DEA analysis on each pair of parameters with correlation efficient rlv ≥ α in and/or , a set of potential energy savings representing the best achievable efficiency of the respective machine/ECU can be obtained. Such results would visualize and quantify the energy inefficiencies, and allow energy managers and management to effectively plan and reduce energy consumptions.
Facilities energy consumption continuously changes together with differences in opening hours and holidays, and the fluctuating outside temperature. Buildings need heating when the weather is cold, and cooling when it is warm. These variables are important to understand energy consumption in buildings.
Models for load forecasting
Load forecasting has several useful applications. First, forecasting may improve the understanding of how energy consumption in a building changes between years. Second, quantification of energy savings from ECMs, and third, detect anomalies. In 1986, the Princeton Scorekeeping Method (PRISM) was introduced as a standard method to measure ECM savings. The PRISM is a simple piece-wise linear regression model with monthly electricity consumption and heating degree-days: as energy data became more available, models using daily and hourly data were proposed, both using multiple linear regression and change-point models. Models that have been established to estimate energy savings that have performed well in comparison with other models are the TVB and CW-GB models.
COMPONENT-WISE GRADIENT BOOSTING WITH PENALISED SPLINES
With an excellent prediction performance within statistics and machine learning the component-wise gradient boosting is robust against multicollinearity and flexible in terms of modelling different types of effects. For a more detailed overview of the applied procedure: We label the outcome variable, energy consumption, y and the predictors (temperature variables and calendar data) x1, …, xp. The objective is to model the relation between y and X : = (x1, …, xp)T, and to estimate the “optimal” prediction of y given x. To achieve this objective, we minimize the loss function ρ(y,f) ∈ over a prediction function f depending on x. Since we use a generalised additive models (GAM) the loss function is the negative log-likelihood function of the outcome distribution. In the gradient boosting the objective is to estimate the optimal prediction function f*, defined by
THE TAO VANILLA BENCHMARK MODEL
The results from the CW-GB model is compared against the TVB model. This model is integrated as a standard load-forecasting model in the commercial software package SAS Energy Forecasting. The model is a multiple linear regression model:
Yt =β0 +β1Mt +β2Wt +β3Ht +β4WtHt +β5Tt
+ β6Tt2 + β7Tt3 + β8TtMt + β9Tt2Mt + β10Tt3Mt
+ β11TtHt + β11Tt2Ht + β11Tt3Ht
where Yt is the load forecast for hour t, βi are the estimated coef- ficients from the least squares regression method; Mt, Wt and Ht are month of year, day of the week and hour of the day. Fur- ther, Tt is the temperature corresponding to time t. Note that the original TVB model includes trend and past loads. In this study the model will reflect how a particular building perform based on a reference period, thus trend and lagged predictors are not included.
Big Data in Smart Energy Management
In a certain sense, smart energy system can be regarded as the convergence of the Internet and the various intelligent devices and sensors spread throughout the energy system. The main source of data is the advanced metering infrastructure (AMI) which deploys a large number of smart meters and other measuring terminals at the end-user side. The smart meters usually collect electricity consumption information every 15 mins, and the meter readings alone have created and accumulated massive amount of data. A large amount of meter reading data will be collected in a distribution network with 1 million metering devices, and the volume of the data can grow exponentially. Data Analytics is changing the way of energy production and the pattern of energy consumption for residential and commercial purposes. Energy big data have brought opportunities in conserving energy needs but typically boil down to a few factors for efficient energy savings in manufacturing companies.
(a) How to effectively Collect, Store and Manage the Energy Big data
The first step to collect and store energy data efficiently is to identify your data sources and needs. You need to know what types of energy data you want to collect, such as electricity, gas, water, temperature, humidity, or occupancy. You also need to know where your data comes from, such as meters, sensors, controllers, or smart devices. Finally, you need to know how often and how much data you need to collect, such as hourly, daily, monthly, or annually, and in what units and formats, such as kWh, m3, °C, or CSV
(b) How to Efficiently Analyze and Mine the Energy Big data
You need to make sure that your data is consistent, complete, and accurate, and that it follows a clear naming and labeling convention. You also need to remove any outliers, duplicates, or missing values that might affect your data quality and analysis. You can use various tools and techniques to organize and clean your data, such as spreadsheets, databases, or data cleansing software.
(c) How to Use the Energy Big Data to Support more Effective and Efficient Decision Makings
Follow the data story and draw conclusion based on energy demand, load shedding. peak demand, power factor and other pointers. Make infromed decisions based off the data analyzed and draw energy savings plans mapping out decisions to run the facility by them.
Big data is still in its infancy, and most of the related big data-driven smart energy management technologies are not mature. With the deepening of scientific research and industrial development, people’s understanding and awareness of smart energy management will also changing.
Other Ways Manufacturing Companies Can Conserve Energy
Strategy 1: Manufacture Using Partially Recycled Materials
In contributing to a more sustainable future by driving development of eco-friendly solution, companies can use manufacturing products that can be reprinted or refurbished, and employ services dedicated to reusing materials for conservation efforts.
Strategy 2: Maintain Equipment to Eliminate Energy Waste
One of the possible causes for energy waste is improper equipment maintenance. Motors may run hot due to alignment issues or damaged bearings, for example. To reduce energy consumption, organizations can regularly inspect their plant’s equipment and devices, determine the most efficient approach for maintenance, upgrades, and replacements and place equipment in low-energy mode during certain hours. Predictive maintenance techniques can help to anticipate the need for equipment maintenance. Knowledge around material degradation, along with data about the processes and equipment, will give your team valuable information about the reliability of your equipment.
Strategy 3: Routinely Monitor Your Processes & Materials
Another way to optimize energy efficiency is to monitor your plant’s processes and materials on a regular basis. This allows for improvement across teams to pinpoint root causes and streamline problem-solving throughout their organization.
These are just a few of many strategies that your organization can implement to achieve energy conservation success. Insights from your data can help you to decide the most appropriate strategies for your business. With a good action plan powered by data-driven decisions, businesses can uncover valuable saving opportunities to reduce their carbon footprint.Data-driven approach were developed for identification of energy saving opportunities and quantification of their respective energy saving potentials. These approach opens a new way to discover hidden energy saving opportunities which are invisible and hard to identify. This will be particularly useful for companies with successful energy improvement programs to identify and plan areas for further improvement.
Cited Works
Kaile Zhou, Chao Fu, Shanlin Yang, Big data driven smart energy management: From big data to big insights, Renewable and Sustainable Energy Reviews, Volume 56,2016, Pages 215–225, ISSN 1364–0321, (https://www.sciencedirect.com/science/article/pii/S1364032115013179)
Bin Song*, Yintai Ao, Li Xiang, K.Y.Ng Lionel, Data-driven Approach for Discovery of Energy Saving Potentials in Manufacturing Factory, https://pdf.sciencedirectassets.com/282173/1-s2.0-S2212827118X00043/1-s2.0-S2212827117309319)
