基于工业控制系统数据对未来设备运行状态和生产过程进行趋势预测，并提前识别潜在故障，是保障工业安全生产和提升生产效率的关键手段。然而，趋势预测算法面临着工控时序数据中复杂的序列内和序列间相关性精准理解难题。为此，提出了一种新的工控时序数据预测模型。首先，在数据预处理时引入小波变换，从时域和频域局部同时捕捉工控时序数据的非平稳的突变、瞬态特征；然后，融合自适应图卷积学习工控时序数据序列间相关性，捕捉数据序列间变化特征；最后采用动态加权的状态空间模型Mamba识别每个时间序列内的多变量的依赖关系。在7个公开数据集以及自建工控时序数据集上进行长时序预测实验表明，相比于7个时序数据预测模型，该模型的MSE降低34%，MAE降低37%，展现了该模型显著的性能优势。

It is an effective means to ensure industrial safety production and improve production efficiency to predict the future trend of equipment operation status and production process based on industrial control system data, and identify potential faults in advance. However, trend prediction algorithms face challenges in understanding the complex correlations within and between sequences in industrial control data. To address this, a time series prediction model for industrial control systems is proposed. The model introduces wavelet transform during data preprocessing to locally capture non-stationary behaviors, abrupt changes, and transient features in both time and frequency domains. It integrates adaptive graph convolution to analyze temporal correlations across sequences, identifying periodic and trend patterns in the data. Finally, a dynamically weighted state-space model, Mamba, is used to uncover multivariate dependencies within each time series. Experiments on seven public datasets and a custom-built industrial control dataset demonstrate that the model reduces MSE by 34% and MAE by 37% compared to seven other prediction models, demonstrating its significant performance advantages.