Inadequate load output diagnosis of ultra-supercritical thermal power units based on MIWOA multi-label random forest

Aiming at the problem of considerable economic losses caused by the large fluctuation range of response load and Automatic Generation Control (AGC) commanded load of ultra-supercritical thermal power units, an inadequate load output diagnosis model of multi-label random forest with multi-improved whale optimization algorithm (MIWOA-MLRF) is proposed. Thermal power units are a kind of high-dimensional, nonlinear, and complex industrial complex, which brings difficulty for conventional mechanistic models in comprehensively analyzing the inadequate output causes of the units. With the help of extensive data analysis and artificial in-telligence algorithms, a multi-label random forest (MLRF) for inadequate output cause analysis is constructed. To improve the accuracy of WOA for MLRF classification, a good point set strategy is used to optimize the initial population distribution of WOA, and an improved convergence factor is used to control the speed of WOA search. Three Gaussian simulation datasets with a different number of features and labels and three sets of thermal power operation data with varying periods are utilized. The test results show that the missing alarm rates (MAR) of MIWOA-MLRF are 0.301, 0.621, and 0.802 under three sets of Gaussian data tests, which are the smallest values among all analytical models. Under three sets of thermal power unit operation data, the average values of MIWOA-MLRF's MAR are 2.2%, 0.2%, and 1.5%, respectively, which have lower missing alarm rates compared with other algorithms. Furthermore, in the false alarm rate (FAR) comparison, the FAR of MIWOA-MLRF de-creases significantly compared to MLRF. In the optimization time comparison, the optimization time required for MIWOA-MLRF is about 1000 s, which saves 33% on average compared to the optimization time of other algo-rithms. The result indicates that MIWOA-MLRF can guarantee the same better real-time performance with reduced FAR and MAR of inadequate output cause classification.