Against the problem that the conventional timed and quantitative soot blowing mode is prone to cause local over-blowing and under-blowing of the waterwall, studies are carried out by relying on effective monitoring methods. As slag deposition on waterwall is a key factor affecting the safe and economic operation of thermal power boilers, long-term unresolved local over-blowing or under-blowing will not only accelerate the corrosion and wear of the waterwall, but also increase energy consumption and operational costs of power plants. Therefore, the core goal of this research is to establish a precise soot blowing algorithm to replace the conventional timed and quantitative soot blowing mode and realize adaptive and efficient soot blowing control.

A new type of waterwall slagging monitoring sensor was used to monitor the in-furnace waterwall surface temperature, which can collect real-time, continuous and high-precision temperature data to lay a reliable foundation for subsequent model construction. Three machine learning methods, including eXtreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM) and random forest regression (RFR), were compared to construct theoretical in-furnace waterwall surface temperature models under clean waterwall conditions, and a calculation method for waterwall slagging factor was proposed. On this basis, a precise soot blowing algorithm was established. To verify the optimization effect of this algorithm, a 3-month practical application test was carried out in a 1 030 MW thermal power unit, and the operation data was compared with the original timed and quantitative soot blowing mode.

The research shows that the theoretical in-furnace waterwall surface temperature model established by the random forest regression method performed the best, with an R² of 0.92, δ_MSE of 73.77, and δ_MAPE of 1.16%. The implementation of the precise soot blowing algorithm is significantly better than the original quantitative soot blowing mode, with a significant reduction in soot blowing frequency and no deterioration of the waterwall slagging state, and the local maximum temperature of the waterwall is controlled within the safe range, avoiding the risk of tube explosion caused by overheating.

This algorithm reduces the consumption of soot blowing steam while ensuring the safety of boiler operation, which directly reduces the daily operation cost of the power plant. Moreover, the reduction of soot blowing frequency also reduces the influence of high-temperature steam on the waterwall, effectively extending the service life of the waterwall and reducing the maintenance cost of the boiler. It can be popularized and applied in thermal power plants of different capacities, and has extremely high application value.