Terra Daily — July 7, 2026

Research Worth Reading

• Physics-Informed Dynamic State Estimation for Current Transformers Using Graph Neural Networks — This paper proposes a physics-informed graph neural network approach to estimate the dynamic state of current transformers under transient core saturation, improving secondary current accuracy. For engineers pivoting to climate, it demonstrates how GNNs can be applied to grid instrumentation where physical constraints matter.
• RCOA Extension and Applications — This work extends the Relaxed Convex Obstacle Avoidance formulation, enabling fully convex optimal control problems for obstacle avoidance with analytical convergence guarantees. Mechanical and control engineers can apply these methods to autonomous navigation for infrastructure inspection, such as offshore wind or solar farms.
• Overload-Based Cascades in Multiplex Flow Networks with Partial Functionality — The study models cascading overload failures in multiplex flow networks where nodes can operate with partial functionality, unlike binary up/down assumptions. This matters for climate engineers building resilience simulations of interdependent power grids and supply chains under stress.
• Data-driven Kernel-based Predictive Control with Stability and Robustness Guarantees — This paper analyzes closed-loop stability and robustness of a data-driven kernel-based predictive control scheme that learns directly from input-output data. For climate tech, such control approaches can be deployed on building HVAC or grid storage without first-principles models, using standard ML tooling.
• Exact Closed-Form Feedforward Inversion for Dual-Bridge Series Resonant DC/DC Converter via State-Plane Analysis — The authors derive exact closed-form feedforward inversion maps for a dual-bridge series resonant DC/DC converter using state-plane trajectory analysis across four modulation variables. This gives power electronics engineers a direct control method for high-efficiency converters in EV charging or renewable energy interfaces.
• Physics-Informed Neural State-Space Modeling of Battery-Electric Vehicle Dynamics for Closed-Loop Automated Parking Simulation — This work presents a physics-informed neural state-space model of a production battery-electric vehicle’s parking dynamics, identified from field-test maneuvers for closed-loop simulation. It shows how hybrid first-principles plus neural modeling can support EV control development, a direct application for engineers moving into transport decarbonization.

Today’s Synthesis

A concrete build for engineers this week: a grid-edge resilience stack that chains three of today’s papers into a working control loop. Start with Physics-Informed Dynamic State Estimation for Current Transformers Using Graph Neural Networks to recover accurate secondary current states under transient core saturation — this runs on PyTorch Geometric and fits directly onto existing substation telemetry without new hardware. Pipe those estimated node states into the Overload-Based Cascades in Multiplex Flow Networks with Partial Functionality simulator, which models partial-functionality degradation instead of binary up/down failure, producing a realistic stress map of interdependent grid and supply-chain links under load. Close the loop with Data-driven Kernel-based Predictive Control with Stability and Robustness Guarantees : train the kernel MPC directly on input-output data from distributed battery storage or building HVAC to shed or shift load when cascade risk crosses a set threshold. The full pipeline uses standard ML and convex optimization tooling (CVXPY, scikit-learn kernels) with no requirement for a first-principles grid model. For a software or ML engineer entering climate work, this is a tractable weekend-scale prototype that exercises GNN inference, network simulation, and certified control in one deployable package.