
關鍵詞:循環冷卻水系統;運行狀態預測;遺傳算法中圖分類號:TM621 文獻標志碼:ADOI:10.19968/j.cnki.hnkj.1003-5168.2025.14.002文章編號:1003-5168(2025)14-0013-04
Research on Operation Status Prediction of the Circulating Cooling WaterSystem in PowerPlantsBased onthe Combinationof LSTM and EMD
LIU Feiyu (Intelligent Control Industry College,Henan Chemical Technician College,Kaifeng 4750Oo,China)
Abstract: [Purposes] To address the insufficient prediction accuracy of the operation status in power plant circulating cooling water system,a prediction method combining Genetic Algorithm-optimized Bidirectional Long Short-Term Memory neural networks (GA-BiLSTM) and Empirical Mode Decomposition (EMD)is proposed.[Methods]EMD is employed to decompose the original data into multiple Intrinsic Mode Function (IMF)components,thereby reducing data complexity.With the help of the Genetic Algorithm(GA),the hyperparameters of the Bidirectional Long Short-Term Memory neural network (BiLSTM) are optimized to improve the performance of the model. The decomposed IMF components are then individuallyfed into theoptimized GA-BiLSTM model for prediction,with final resultsobtained through reconstruction.[Findings] Experimental results demonstrate that all prediction error metrics of this model remain at low levels,with a 55% improvement in prediction accuracy compared to conventional models.[Conclusions] The prediction method based on the combination of LSTMand EMD can provide strong assurance for stable operation of the circulating cooling water system in power plants.
Keywords: circulating cooling water system; operation status prediction; Genetic Algorithm
0 引言
電廠循環冷卻水系統在工業生產中占據關鍵地位,其穩定運行對保障設備正常運轉、提高生產效率和降低能耗具有重要意義。然而,電廠循環冷卻水系統的運行過程受多種復雜因素影響,具有高度的非線性和不確定性,傳統預測方法難以滿足其高精度預測需求。
近年來,隨著深度學習技術在預測領域的不斷深入研究,其中長短期記憶神經網絡(LongShort-TermMemory,LSTM)因其獨特的門控機制,能有效處理時間序列數據中的長期依賴問題,在工業預測等領域得到廣泛應用[。但LSTM網絡在超參數選擇過程中對模型的性能要求太高,若設置不合理將會出現參數過度擬合或擬合欠缺的問題。而遺傳算法(GeneticAlgorithm,GA)作為一種高效的全局優化算法,能夠在搜索空間中快速找到較優解,可用于優化LSTM網絡超參數[2]。同時,經驗模態分解(EmpiricalModeDecomposition,EMD)能將復雜的時間序列數據自適應分解為多個具有不同特征尺度的本征模態函數(IntrinsicModeFunction,IMF)分量,使數據特征更加清晰,有助于提升預測模型的精度[3]。因此,本研究將GA、雙向長短期記憶神經網絡(Bidirectional Long Short-Term Memory,BiL-STM)和EMD相結合,提出一種新的預測模型,有望實現對電廠循環冷卻水系統運行狀態的準確預測。
1相關算法原理
1.1雙向長短期記憶神經網絡(BiLSTM)
雙向長短期記憶神經網絡BiLSTM是在LSTM基礎上發展而來的一種神經網絡結構。通過設置輸入、遺忘和輸出門控制機制,有效解決了信息計算過程中的梯度消失和爆炸問題,能夠更好地捕捉時間序列中的長期依賴信息。……