Multi-field Coupling Precision Control Technology of Downhole Heater for Oil-Rich Coal In-Situ Pyrolysis

Targeting the extreme downhole environments of oil-rich coal in-situ pyrolysis—characterized by high pressure up to 21MPa, severe corrosion with 3% H₂S content, substantial temperature fluctuations from 120℃ to 650℃, and inherent difficulties in data acquisition—traditional heater control systems suffer from inadequate precision, weak anti-interference capability, poor adaptability, and lack of effective life prediction methodologies for unoperated scenarios. To address these issues, this study conducts in-depth research on an intelligent control system centered on the full-link closed-loop control logic of "perception-decision-execution-feedback".

A novel three-model collaborative decision-making architecture integrating "physical benchmark-condition adaptation-time series supplement" is established. The Weibull model serves as the physical life baseline to ensure compliance with equipment aging laws, reflecting the late-stage accelerated aging characteristic of downhole heaters with a shape parameter greater than 1, a scale parameter representing the characteristic life corresponding to a 63.2% failure probability, and a position parameter defining the minimum safe life threshold. The XGBoost model quantifies the impact of operating conditions such as pressure and corrosion rate through an additive integration mechanism, enabling accurate life correction without relying on the target equipment’s own operating data. The LSTM network captures time-series dynamic anomalies via its gate control unit structure, and its weight is adaptively reduced in unoperated scenarios while the weights of the preceding two models are enhanced through a dynamic weighted fusion approach, addressing the limitation of single-model dependence on operational data.

A hierarchically collaborative control architecture is designed. The perception layer deploys a high-temperature and high-pressure resistant sensor array, achieving 10ms-cycle data transmission through the PROFINET industrial bus and MQTT/OPC UA dual protocols to mitigate environmental interference. The decision layer integrates adaptive fuzzy control with PID regulation, interfaces with a digital twin system for real-time state mapping and fault pre-diagnosis, and embeds a predictive maintenance model based on resistance drift rate and thermal response time. The execution layer takes Siemens PLC as the core, complemented by a thyristor regulator with a response time of less than 10ms and an independent hard-wired safety loop that terminates power supply within 0.5s under extreme conditions. The feedback layer calibrates PID parameters every 10 days and optimizes model weights for full-cycle iterative optimization.

Hardware optimization involves integrating PLC with high-temperature resistant sensors and developing reliable packaging processes including wellhead sealing and cable crossing sealing. Validated by 1200-hour continuous operation on a 205m-deep in-situ test platform, the system achieves life prediction accuracy with a coefficient of determination of no less than 0.9 and a mean absolute percentage error of no more than 10%, controls outlet temperature fluctuation within ±3℃, maintains a control response time of no more than 10ms, an insulation resistance of no less than 100MΩ, and a thermal efficiency of 89.7%. Stable performance is retained even with 30% data loss, providing a systematic theoretical and engineering framework for the safe and efficient operation of oil-rich coal in-situ pyrolysis equipment.