Data Analysis in the Energy Monitoring System Georgii Mikriukov 1 Perm National Research Polytechnic University Electrical engineering faculty Perm – Köhten
2 GridVis – Data acquisition software OpenJEVis – Open source software, basis of energy monitoring system 2 Energy monitoring system
Functions of DAS DAS – Data Analysis System Primary data processing Data smoothing Statistical parameters calculation Energy consumption prediction Atypical consumption detection 3 3
DAS functioning algorithm 4 E – raw data E’ – filtered data E* – smoothed data S – statistical data Ep – predicted energy consumption ci – clusters centroids ti – consumption time intervals fi – function of consumption ΔE = E’ – E* 4
5 Smoothing (moving average) 2N+1 – number of smoothing points (3, 5 or 7) x i – raw data point xs i – smoothed data point Atypical consumption detection (ACD) ΔE < 0 – equipment switching off ΔE > 0 – equipment switching on 5 DAS functions: smoothing & ACD
6 Clusterization (k-means) V – standard deviation k – number of clusters x j – raw data point S i – cluster c i – centroid Classification 6 DAS functions: clusterization & classification
7 Statistical parameters maximal consumption average consumption (per day, week, month etc.) leveled consumption value (part of maximal) Working time in different modes Energy consumption prediction 0,75P av < P di <1,25P av – normal consumption P di < 0,75P av – lowered consumption P di > 1,25P av – heightened consumption P di – average consumption per day P av – average consumption for the whole period 7 DAS functions: statistics & prediction
8 Data analysis 8 Period of time: Statistical parameters Maximal consumption (Pmax): 2049,67 W· h Average consumption (Pav): 352,03 W· h Laboratory workload (P > 0,1Pmax): 54,91% Energy consumption, W · h
9 9 Atypical consumption detection ΔE < –600 – emergency event probability E’ &E*, W · h Δ E, W · h Period of time: Filtered (E’) and smoothed (E*) data ΔE = E’ – E* Period of time:
3 main consumers: lighting (70%) lighting (30%) laboratory stands 10 Data clusterization 2 3 =8 combinations
11 Energy consumer classification Class Lighting 70% Lighting 30% Standscici t1t1 t2t , , , , , , , ,14266 c i – cluster centroid (Y coordinate) 11
12 Energy consumption prediction weeks 3 weeks (more accurate) consumption, W · h Period of time:
Conclusion Functions and structure of DAS has been considered Statistics of one month ( ) has been analyzed Energy consumers were classified according to clusterization data Prediction curve of energy consumption has been calculated
Thank you for attention 14