Internet Cable Path Planning - Overcoming Challenges of Excess Data Availability or Missing Ground Motion Data

Description

Submarine telecommunication cables play a crucial role in the internet infrastructure. They cost billions of US dollars and carry 99% of international data. Their breakage (due to e.g. earthquake) can cause internet shutdown with grave socio-economic consequences. To achieve the minimization of cost along with cable breakage risks, under our successfully completed RGC/CRF project during 2014-2017, we have developed a new automated methodology, based on real data, for cable path planning on the earth's surface modeled by a 2D manifold in 3D space. Current industry practice is to do cable path planning manually. As the length of long-haul cables may be thousands of kilometers, manual path planning without a scalable automated software-tool is costly and may not achieve the desired optimal cost-risk tradeoff. Our close collaboration with our industry partners enabled us to consider all relevant practical design criteria in our path planning methodology. This project will optimize cable path planning considering multiple criteria of both cost and risk. It has the following two key novel aspects. 1. Cable path planning optimization considering available dense data on topography, geology, and ground motion. 2. Seismic hazard map development for cable path planning by complementing seismic data in relevant target areas where ground motion data is unavailable.  We will overcome the computationally prohibitive massive data size. First, we will use a multi-resolution scaling approach, starting from low-resolution gridding and suitably spatially averaged data and iteratively to find the optimal path. Second, we will use parallelization to implement our existing provably optimal algorithms and load the data into the computer. We will develop a new distributed shortest path algorithm based on the belief propagation method and demonstrate its advantages in terms of running times and solution quality over existing methods. This will result in optimal path planning over ultra-long-distance for a network with billions of nodes. The optimal solutions achieved by parallel processing will help assess and calibrate the multi-resolution algorithms. To develop high-quality seismic hazard maps for cable path planning with limited data availability, we will use machine learning algorithms to learn the relationship between topography and seismic information from areas where both are available to infer hazard maps in areas where information is limited.  We will develop solutions for both single cable path planning and a network of cables toachieve optimal tradeoffs between cost and risk and will continue to work closely withour industry partners to achieve industry standardization and impact.