In the actual oilfield environment, the complex and mutable situations encountered in production still face significant challenges to traditional pumpjack operation mode, which heavily relys on expert experience and is prone to major decision-making errors. In the meantime, as the petroleum industry continues to evolve, a vast amount of fragmented knowledge with loose structure has been accumulated in the field of pumpjack operation. Therefore, in order to enhance the efficient utilization of internal knowledge within oilfield, this study takes the reasonable working system as its theme to sort out the dynamic liquid level monitoring, indicator diagram analysis, determination of reasonable submergence degree and other controlling factors. Then, this study proposes a decision support model that integrates the pattern matching ReteOO algorithm with the knowledge graph. By using an improved LSTM-CRF model and HanLP tool, key information was successfully extracted from the domain dataset; and validation was conducted using metrics such as precision, recall, f1-score, and area under curve (AUC). Research results demonstrate that the entity recognition rates on domain-specific datasets reached 88.2%, 90.1%, 89.2% and 89.1% respectively. Thereby, a complete knowledge system of reasonable working system for the pumpjack domain was constructed. Finally, the effectiveness of the decision model in optimizing the reasonable working system of the pumpjack well is demonstrated through several examples, providing strong scientific support for the fine management and operation of the pumpjack.