Abstract:Industrial time-series, as a structural data responds to production process information, can be utilized to perform data-driven decision-making for effective monitoring of industrial production process. However, there are some challenges for time-series forecasting in industry, e.g., predicting few-shot caused by data shortage, and decision-confusing caused by unknown treatment policy. To cope with the problems, we propose a novel causal domain adaptation framework, Causal Domain Adaptation (CDA) forecaster to improve the performance on the interested domain with limited data (target). Firstly, we analyze the causality existing along with treatments, and thus ensure the shared causality over time. Subsequently, we propose an answer-based attention mechanism to achieve domain-invariant representation by the shared causality in both domains. Then, a novel domain-adaptation is built to model treatments and outcomes jointly training on source and target domain. The main insights are that our designed answer-based attention mechanism allows the target domain to leverage the existed causality in source time-series even with different treatments, and our forecaster can predict the counterfactual outcome of industrial time-series, meaning a guidance in production process. Compared with commonly baselines, our method on real-world and synthetic oilfield datasets demonstrates the effectiveness in across-domain prediction and the practicality in guiding production process

What problem does this paper attempt to address?

The paper primarily addresses the challenges present in industrial time series forecasting by proposing a novel solution. Specifically, the paper attempts to solve the following key issues: 1. **Problems caused by data scarcity**: In industrial production processes, collecting and labeling sufficient time series data to train models is very expensive and sometimes even impossible. This leads to "cold-start" and "few-shot" problems in the target domain. 2. **Lack of causal guarantees**: Although there is a wealth of theoretical analysis for industrial time series forecasting, these analyses encounter substantial obstacles in the causal relationships within time series that include treatment strategies. Causal relationships can provide intrinsic links between multiple and multivariate time series, so obtaining causal guarantees is crucial for improving forecasting performance and providing production guidance. 3. **Characterizing cross-domain invariance from causal relationships**: To effectively handle data across different domains, it is necessary to construct a time-sensitive model that can use attention mechanisms as an encoder to build cross-domain invariant representations. However, traditional attention mechanisms reconstruct representations based on correlation rather than causality, leading to a lack of causality during domain transfer, making it difficult to estimate the causal impact of policies. To address the above issues, the paper proposes a framework called **Causal Domain Adaptation (CDA)**. CDA combines the ideas of adversarial training and causal inference to construct an industrial time series predictor aimed at achieving causal modeling across treatment-outcome sequences and policy sequences. Its main contributions include: - **Time-varying treatment-invariant representations**: Unlike traditional attention mechanisms, CDA uses counterfactual reasoning to construct domain-invariant representations across time, i.e., treatment-invariant information. This information can break the association between historical matching and treatment alignment. Thus, even with limited data, it can provide sufficient temporal information for time series modeling. - **Counterfactual estimation of future production activities**: To estimate counterfactual outcomes under treatment policies, CDA integrates Conditional Average Treatment Effect (CATE) estimation into the sequence-to-sequence architecture. In this way, CDA can answer the question of which policy is most effective for production activities and demonstrates its application value in the oil and gas sector. The experimental section showcases CDA's performance on real-world oil datasets, achieving good results not only in monthly oil production forecasting but also in selecting the optimal treatment policy to improve oil production.

Domain Adaptation for Industrial Time-series Forecasting via Counterfactual Inference

Transferable Time-Series Forecasting under Causal Conditional Shift

A Domain Adaptation Framework for Short-term Traffic Prediction.

Pre-training enhanced unsupervised contrastive domain adaptation for industrial equipment remaining useful life prediction

Causal Domain Adaptation with Copula Entropy based Conditional Independence Test

Source-Free Domain Adaptation for Transferable Remaining Useful Life Prediction of Machine Considering Source Data Absence

A Fine-Grained Adversarial Network Method for Cross-Domain Industrial Fault Diagnosis

Causal Mechanism Transfer Network for Time Series Domain Adaptation in Mechanical Systems

Multi-level deep domain adaptive adversarial model based on tensor-train decomposition for industrial time series forecasting

Time-series domain adaptation via sparse associative structure alignment: Learning invariance and variance

Deep causal representation learning for unsupervised domain adaptation

Dual Cross-Attention Transformer Networks for Temporal Predictive Modeling of Industrial Process

Feature structured domain adaptation for quality prediction of cross working conditions in industrial processes

Scalable Causal Domain Adaptation

Dual Structural Consistent Partial Domain Adaptation Network for Intelligent Machinery Fault Diagnosis

Few-shot Domain Adaptation by Causal Mechanism Transfer

Multisource Domain Feature Adaptation Network for Bearing Fault Diagnosis Under Time-Varying Working Conditions

LLM-Enhanced Causal Discovery in Temporal Domain from Interventional Data

Adversarial Regressive Domain Adaptation Framework for Infrared Thermography-based Unsupervised Remaining Useful Life Prediction

Domain Generalization in Time Series Forecasting