Tianquan Solver: Smart Brain for Major Grid System

Dubbed as the "core of industrial software," solvers are a key tool for finding solutions to complex mathematical programming problems. They are essential for deriving optimal operation solutions in sectors like energy, finance and logistics, according to Peng Chaoyi, technical lead of Tianquan at the Electric Power Dispatching and Control Center, China Southern Power Grid Co., Ltd. (CSG).

Developed by CSG, the Tianquan Solver is 14 percent faster than imported solvers in terms of computing performance. It has, to date, supported efficient computation for over 7,000 model nodes and more than two million clearing variables within the southern China regional electricity market (Guangdong, Guangxi, Yunnan, Guizhou and Hainan provinces and regions) since the market initiated continuous settlement.

Laying the foundation

An electricity pricing reform took place in 2015, and solvers became key to finding the most reasonable solution. Domestic solvers lagged behind at that time, and CSG decided to catch up.

Many domestic solvers used open source underlying codes that were not optimized, Peng said, adding that this was the biggest reason for the lagging performance.

To tackle this problem, the research team consulted experts from dozens of institutions in mathematics, computer science and electricity. They compiled vast experience, then abstracted these operational insights and physical principles into mathematical models, generating languages that solvers can understand.

In Peng's opinion, the biggest challenge was to find the balancing point and enhance the precision of "translating" the dispatching and operation experience. "Some of the experience was accumulated for years, and the decision needed to be split into different models without being too complicated," he said.

Based on comprehensively understanding the experience and back-and-forth discussions, Tianquan achieved the integration of power system expertise into the solution calculation process for the first time.

Practical test

However, getting the codes right was only the first step before Tianquan could make its mark in the power market. It took over 20,000 seconds (over five hours) for the solver to process its first case, and that was 11 times as long as imported solvers took, Zhou Huafeng, senior manager at the automation department of Electric Power Dispatching and Control Center, CSG, said.

To optimize Tianquan, the researchers first went through mountains of research papers and materials, but the approaches they found could only achieve small-scale optimization, which was almost ineffective for industrial-level applications.

They went back to the prototype of Tianquan, and made it do a large amount of case calculating.

The researchers then broke down and analyzed the entire calculating process, looked for the reason of slow operation in the codes line by line, and cobbled together the answers.

In addition, each member of the research team contributed personal dispatching operation experience and specific expertise into the algorithms of Tianquan, deepening the optimization of its underlying codes.

After countless simulations, Tianquan's calculating speed became 14 times faster than that of imported solvers, when they solved the same case.

For general solvers, there is certain room for error in their application fields. However, the reliability of Tianquan directly impacts grid security, affecting numerous operational units and power consumers, with virtually zero margin for error, according to Liang Yanjie, technical expert at the Electric Power Dispatching and Control Center, CSG.

To ensure accuracy, the team formed cases by collecting real data from a lot of power plants and transformer substations of each level of power grid within CSG. These cases were used by Tianquan for practice during its "internship," and each case contained hundreds of thousands of clearing variables.

After the "internship," Tianquan advanced to the frontline. It dealt with real-time data and continuously optimized each parameter setting, however its calculation results were not adopted.

Undertaking key tasks

With more real practices, the processing efficiency of Tianquan began to rival that of imported solvers, and its results continue to be optimized.

In 2022, as Tianquan began to be dispatched in parallel with imported solvers, questions soon emerged from various departments.

The team, however, was unfazed. "Every step of Tianquan's algorithm is well-founded, and we can clearly explain each computational process to the executing units," Peng said, adding that the self-developed solver can demonstrate the underlying logic for each problem.

The electricity market never sleeps, and Peng and colleagues stayed ready to answer any question, and continuously optimized Tianquan based on real situations. Finally, Tianquan's calculation results were recognized by the organizations using the solver.

After nearly two years of parallel operation with imported solvers, Tianquan has officially transitioned to independent operation as the primary solver. Its operational support expanded from two days to the entire month, and its coverage extended from traditional energy to new energy sources.

Following 35 months of preparation and 12 rounds of internal testing, Tianquan ultimately achieved independent support for continuous settlement in the southern China regional electricity market.

Ge Dongdong, professor at the Institute for Intelligent Computing at Shanghai Jiao Tong University, believes this achievement resolves the technical challenges of clearing optimization in ultra-large-scale electricity markets, reaching internationally leading standards.

Source: Science and Technology Daily