To foster in-depth dialogue and integration between academia and industry, on April 23, Dr.Xiaohua Zhou, Founder and CEO of DolphinDB, was invited to Southeast University to share insights on the topic "High-Frequency Data Processing and Algorithm Implementation in Next-Generation Power Systems."

The lecture hall was filled to capacity. Professor Xu Qingshan, Associate Dean of the School of Electrical Engineering, served as the opening host. Taking the stage, he joked: "Power market reform has entered deep waters, and AI applications are no longer just theoretical. Today's lecture may very well change the way you think about power data."

With these words, the atmosphere was instantly energized. So what exactly did Dr.Xiaohua Zhou discuss? Below is a quick recap of the lecture highlights—

Dr. Zhou did not start with abstract concepts but instead posed a scenario: At 9 p.m., when off-peak electricity rates take effect, electric vehicles across the city begin charging simultaneously—can the power grid handle the load?

Everyone knows the answer: It cannot. But how can this be resolved? Through intelligent staggered charging to smooth loads and price leverage to guide consumer behavior.

This brings us to two key concepts in new power systems: balance and pricing. In the past, these relied on planning; today, they rely on data.

Why are traditional approaches no longer effective?

Conventional solutions adopt a technical route of "multi-component layered assembly," with data scattered across multiple systems. Cross-system correlation analysis depends on ETL and manual alignment of data definitions. Time-series data and business data are forcibly stored separately: time-series databases offer fast queries but weak cross-table joins, while relational databases are convenient for modeling but cannot handle massive streams of curve data and high-frequency writes.

More critically, real-time and offline processing remain siloed. Source-side deviation analysis requires real-time power alongside historical forecasts; load-side response verification requires real-time load paired with baseline models; dispatch rolling validation depends on real-time status combined with topology constraints. These scenarios cannot be correlated, making complex analysis difficult to execute. Furthermore, a single business scenario often spans multiple computing engines, and any change in rules requires code modification, redeployment, and reconciliation, leading to increasingly complex systems.

DolphinDB's Approach: Power Data Foundation + Real-Time Computing Engine

Positioned as both a power data foundation and a real-time computing engine, DolphinDB covers the entire data pipeline — ingesting from sources such as SCADA systems, power acquisition platforms, weather feeds, and trading settlements via an API gateway, and processing through distributed storage, stream computing, CEP engines, rule engines, and machine learning modules, all backed by over 2,000 built-in functions. The result is a unified platform supporting four key scenarios: source-side monitoring, grid-side perception, load-side aggregation, and trading settlement.

Four Scenarios, Four Practical Deployments

• Source-Side Curve Governance: A single province has approximately 30 million measurement points, with data reported every 15 minutes. Issues include abnormal values, backward spikes, and missing data. Traditional solutions using RDS and Java take several hours to process. DolphinDB consolidates anomaly detection, data fitting, and multi-dimensional aggregation entirely within the database, completing the task in minutes.
• Grid-Side Vibration Monitoring: Main transformer vibration sampling reaches as high as 50 kHz. Traditional approaches either upload all data (overwhelming bandwidth) or use fragmented edge programs that are difficult to maintain. DolphinDB performs FFT and anomaly detection at the edge with millisecond-level response, while fully retaining abnormal waveforms.
• Load-Side Demand Response: Adjustable loads participate in peak shaving and valley filling. Traditional solutions using relational databases and Spark suffer from slow queries and poor real-time performance. DolphinDB builds a unified foundation, enabling closed-loop integration from baseline construction and real-time monitoring to subsidy settlement.
• Trading-Side Settlement Simulation: Monthly settlement rules change frequently. Traditional solutions require code changes and redeployment, resulting in slow iteration. DolphinDB supports parameterized settlement, allowing rule changes to be recalculated in seconds, ensuring traceability for every transaction and enabling what-if simulations before policy adjustments.

DolphinDB's Technical Foundation

• Distributed Storage: Native distributed architecture with PAX hybrid row-column storage achieving high compression ratios, supporting ACID strong consistency.
• Multi-Model Storage Engine: DolphinDB's multi-model storage capability is built on the TSDB engine, achieving efficient device point management through structures such as point latest value cache tables and point static information tables. Combined with the IOTANY data type, same-type point data can be stored centrally, significantly improving write and query efficiency.
• 2,000+ Built-in Functions: Covering mathematics, statistics, financial analysis, and other scenarios, with support for plugins and module extensions.
• 20+ Streaming Engines: Covering common real-time processing scenarios such as stateful computation, window aggregation, anomaly detection, and multi-source heterogeneous data correlation.
• Rule Engine: Enables monitoring with automatic rule matching, online updates without business interruption, and millisecond-level latency.
• CEP Engine: Complex event processing supporting event distribution, matching, and callback function execution.

What Can AI Do?

Dr. Zhou then turned to the intersection of AI and time-series databases, walking through several capabilities DolphinDB has already deployed in production.

These include an enterprise-grade RAG agent capable of rapid document retrieval and accurate question answering, as well as VectorDB — a billion-scale vector store with millisecond-level approximate search that, paired with text search, powers a complete knowledge Q&A pipeline.

On the engineering side, the LibTorch plugin allows users to load and run deep learning models directly within DolphinDB — eliminating the need to move data in and out for inference. For compute-intensive workloads, the CPU-GPU heterogeneous computing platform Shark brings the full power of GPU acceleration to bear within the same environment.

Dr. Zhou closed the section with a single line: "Enabling AI to understand time-series data is the fundamental capability of the next-generation data foundation."

Q&A Session

The Q&A session drew enthusiastic participation, with Dr. Zhou and Associate Dean Xu Qingshan fielding a range of technically rigorous questions from students. Topics ranged from millisecond- and microsecond-level data analysis in electromagnetic relay protection to real-time processing at sampling frequencies up to 25,000 Hz — a testament to the depth of technical interest in the room.

From the scenario of EV charging, to governance of tens of millions of measurement points, 50 kHz vibration analysis, demand response closed-loop systems, and the integration of AI with time-series databases, the lecture built progressively—featuring both accessible real-world examples and in-depth industry frontier insights.

Many students noted that the lecture provided a clearer understanding of the data technologies behind new power systems and sparked new ideas for their own research. This deep dialogue between academia and industry not only bridged the gap between classroom learning and engineering practice but also opened a window for students into the industry frontier.

DolphinDB will continue to visit universities, sharing more real-world cases and technical practices, and helping more young professionals master the core competencies needed in the era of high-frequency data.

About DolphinDB

DolphinDB is a high-performance distributed time-series database. Beyond efficient storage and querying of massive datasets, DolphinDB pioneers a fully featured programming language for complex analytics, alongside a high-throughput, low-latency, and developer-friendly stream processing framework — making it one of the most computationally powerful database systems available. DolphinDB is widely adopted by leading financial institutions across securities, fund management, banking, and insurance, as well as top enterprises in IoT-driven industries including energy, power, and industrial manufacturing — significantly enhancing large-scale data analysis efficiency and substantially reducing development costs.

DolphinDB Cerulean Campus Program

To advance university collaboration, DolphinDB has officially launched the Cerulean Campus Program — an initiative dedicated to bridging industry and academia through collaborative innovation and joint talent development. The program aims to integrate DolphinDB into university curricula and cultivate a new generation of fintech professionals equipped with global perspectives, innovative thinking, and hands-on expertise. Collaboration formats include guest lectures, curriculum development, talent training, and joint research, offering students access to rich learning resources, internship opportunities, and research topics.

To date, DolphinDB has established partnerships with a number of leading universities across China, including Shanghai Jiao Tong University, Peking University, Zhejiang University, Fudan University, Nanjing University, Renmin University of China, University of Science and Technology of China, The Chinese University of Hong Kong (Shenzhen), and more.