Time-Series Database (TSDB)

TSDBs are engineered for append-only write workloads, where new data arrives as a continuous stream of time-stamped measurements. Unlike general-purpose databases, they bypass traditional B-tree indexing in favor of log-structured merge trees (LSM) or custom time-partitioned storage engines. This design ensures that a PMU streaming 60 synchrophasor frames per second can write directly to disk without contention, maintaining consistent sub-millisecond write latencies even under sustained ingestion of millions of data points per second.

To manage the massive storage footprint of high-resolution grid telemetry, TSDBs implement automated downsampling and tiered retention policies. Raw 60 Hz PMU data may be retained at full fidelity for only a few weeks, while automatically aggregated 1-minute or 1-hour rollups are preserved for years. This process, often called continuous aggregation, pre-computes statistical summaries (min, max, mean, standard deviation) on ingest, allowing operators to query years of historical oscillation trends without scanning petabytes of raw data.

Synchrophasor data exhibits high cardinality—thousands of PMU device IDs, each with multiple measurement channels (voltage magnitude, phase angle, frequency, ROCOF). TSDBs store data in a columnar format, grouping values from the same measurement channel together. This enables extreme compression ratios (often 10x-50x) using algorithms like delta-of-delta encoding and Gorilla XOR compression, which exploit the fact that time-series values change predictably between consecutive timestamps. A voltage magnitude reading of 1.02, 1.021, 1.019 compresses to near-zero storage cost.

TSDBs expose query languages optimized for time-range scans and windowed aggregations, not arbitrary transactional lookups. Functions like time_bucket(), LAG(), and HISTOGRAM() allow a reliability coordinator to instantly retrieve the average frequency deviation across a 200-bus interconnection during a specific 500-millisecond disturbance window. This is essential for event replay and forensic analysis, where engineers must correlate a transient instability across dozens of PMU streams simultaneously, reconstructing the exact sequence of events that led to a near-miss.

Real-world PMU data streams are imperfect—packets are lost, GPS clocks drift, and PDCs introduce latency jitter. TSDBs provide native functions for linear interpolation, last observation carried forward (LOCF) , and step interpolation to fill gaps in query results without manual data wrangling. When a modal analysis algorithm expects a uniform 30 Hz stream but a PMU dropped three consecutive frames, the TSDB can seamlessly fill those missing timestamps, ensuring downstream stability assessment tools receive a clean, continuous signal.

TSDBs organize data using tags (key-value metadata pairs) rather than rigid relational schemas. A synchrophasor measurement is tagged with attributes like substation=Maple Ridge, voltage_level=500kV, pmu_id=MR-07, and channel=VA. This dimensional model allows operators to query across arbitrary tag combinations—for example, comparing the voltage phase angle on all 500 kV buses in the Pacific Northwest corridor during a specific oscillation event—without pre-defining table joins or complex foreign key relationships.

Specialized algorithms essential for managing the massive volume of PMU data, which can exceed terabytes per day for a single utility. TSDBs use lossless and lossy compression to reduce storage costs without sacrificing analytical fidelity.

• Delta-of-delta encoding: Stores only the change in the rate of change
• Gorilla compression: A Facebook-developed method optimized for time-series floats
• Deadbanding: Discards values that fall within an insignificant change threshold

The automated process of reducing high-resolution raw data into lower-granularity summaries for long-term retention. A TSDB continuously computes aggregated views to enable fast queries over years of data.

• 1-minute averages from 30Hz raw PMU streams
• Max/Min/StdDev rollups for oscillation analysis
• Automatic retention policies that expire raw data after a set period

A critical TSDB capability that allows engineers to query and replay a high-fidelity historical window surrounding a grid disturbance. This turns the database into a 'time machine' for forensic investigation.

• Reconstructs the exact system state before a blackout
• Feeds historical data into Hardware-in-the-Loop (HIL) simulators
• Validates the performance of protection schemes like SIPS

Modern TSDBs extend standard SQL with continuous query semantics, allowing algorithms to process data in-flight before it is written to disk. This enables real-time situational awareness dashboards.

• Sliding window aggregations for ROCOF calculation
• Pattern matching to detect forced oscillations as they occur
• Materialized views that update incrementally with each new PMU frame