基于LSTM预测与泵阀协同的瓦斯抽采主管路智能调控系统构建与应用

Construction and Application of Intelligent Regulation System for Main Gas Drainage Pipeline Based on LSTM Prediction and Pump-Valve Coordination

  • 摘要: 【目的】针对煤矿瓦斯抽采主管路中抽采泵与电动阀门强非线性耦合、瓦斯涌出剧烈波动导致纯流量跟踪滞后与控制目标失配的问题,【方法】提出一种融合长短时记忆网络(LSTM)趋势预测与“先阀后泵”分层协同的智能调控方法,构建了“状态感知-趋势预测-协同决策-动态执行”闭环调控框架;LSTM模型以抽采纯流量、阀门开度、泵频率、抽采负压为输入,实现未来3 min纯流量的滚动预测,采用优先调节阀门、必要时调节泵频率的协同策略,设计了基于连续调节失效触发的目标值动态更新机制,并以九里山矿南风井瓦斯抽采泵站主管路监测调控开展了对比试验。【结果】结果表明:在瓦斯涌出剧烈波动阶段,泵阀协同智能调控的最大动态偏差为±6.2 m3/min,平均绝对百分比误差(MAPE)为6.8%,较人工经验调控和传统PID控制分别降低63%和45%(p<0.01);系统平均响应时间缩短至45 s,泵变频动作次数由12.4次/24 h降至6.1次/24 h,单位抽采纯流量电耗降低18.7%。动态目标更新机制可在抽采达标约束下在线重构控制目标。该方法为煤层瓦斯抽采系统主管路调控提供了高精度、低冲击的自适应控制手段。

     

    Abstract: Objective To address the problems of strong nonlinear coupling between extraction pumps and electric valves in coal mine gas drainage trunk pipelines, as well as the lag in pure gas flow tracking and control target mismatch caused by severe fluctuations in gas emission, Methods an intelligent regulation method integrating Long Short-Term Memory (LSTM)-based trend prediction with a hierarchical "valve-first, pump-second" collaborative control strategy was proposed. A closed-loop regulation framework consisting of state perception, trend prediction, collaborative deci-sion-making, and dynamic execution was established. The LSTM model utilized pure gas flow rate, valve opening, pump frequency, and extraction negative pressure as inputs to achieve rolling pre-diction of pure gas flow over the next 3 min. A collaborative control strategy prioritizing valve adjustment and employing pump frequency regulation when necessary was developed, together with a dynamic target updating mechanism triggered by consecutive regulation failures. Compar-ative experiments were conducted based on the monitoring and control data of the gas drainage trunk pipeline at the Nanfeng Well pumping station of Jiulishan Mine. Results During periods of severe gas emission fluctuations, the maximum dynamic deviation of the pump-valve collaborative intelligent regulation was ±6.2 m3/min, while the mean absolute percentage error (MAPE) was 6.8%, representing reductions of 63% and 45% compared with manual experience-based regulation and conventional PID control, respectively (p<0.01). The average response time was reduced to 45 s, the frequency conversion actions of the extraction pump decreased from 12.4 to 6.1 times/24 h, and the unit energy consumption per pure gas flow decreased by 18.7%. The dynamic target up-dating mechanism enabled online reconstruction of control targets under gas drainage compliance constraints. Conclusions: The proposed method provides a high-precision and low-impact adaptive control approach for the regulation of coal mine gas drainage trunk pipelines.

     

/

返回文章
返回