In order to approximate the underlying process of temporal data, time series prediction has been a hot research topic for decades. Developing predictive models plays an important role in interpreting complex real-world elements. With the sharp increase in the quantity and dimensionality of data, new challenges, such as extracting deep features, recognizing deep latent patterns have emerged, demanding novel approaches and effective solutions. Deep learning, composed of multiple processing layers to learn with multiple levels of abstraction, are now commonly deployed for overcoming the newly arisen difficulties. This paper reviews state-of-the-art developments in deep learning for time series prediction. Based on modeling for the perspective of conditional or joint probability, we categorize them into discriminative, generative and hybrids models. Experiments are implemented on both benchmarks and real-world data to elaborate the performance of representative deep learning-based prediction methods. Finally, we conclude with comments on possible future perspectives and ongoing challenges with time series prediction.

As one of the most essential sources of energy, byproduct gas plays a pivotal role in the steel industry, for which the flow tendency is generally regarded as the guidance for planning and scheduling in real production. In order to obtain the numeric estimation along with its reliability, the construction of prediction intervals (PIs) is highly demanded by any practical applications as well as being long term for providing more information on future trends. Bearing this in mind, in this article, a hierarchical granular computing (HGrC)-based model is established for constructing long-term PIs, in which probabilistic modeling gives rise to a long horizon of numeric prediction, and the deployment of information granularities hierarchically extends the result to be interval-valued format. Considering that the structure of this model has a direct impact on its performance, Monte-Carlo search with a policy gradient technique is then applied for reinforcement structure learning. Compared with the existing methods, the size (length) of the granules in the proposed approach is unequal so that it becomes effective for not only periodic but also nonperiodic data. Furthermore, with the use of parallel strategy, the efficiency can be also guaranteed for real-world applications. The experimental results demonstrate that the proposed method is superior to other commonly encountered techniques, and the stability of the structure learning process behaves better when compared with other reinforcement learning approaches.

The flow variation tendency of byproduct gas plays a crucial role for energy scheduling in steel industry. An accurate prediction of its future trends will be significantly beneficial for the economic profits of steel enterprise. In this paper, a long-term prediction model for the energy system is proposed by providing an adaptive granulation-based method that considers the production semantics involved in the fluctuation tendency of the energy data, and partitions them into a series of information granules. To fully reflect the corresponding data characteristics of the formed unequal-length temporal granules, a 3-D feature space consisting of the timespan, the amplitude and the linetype is designed as linguistic descriptors. In particular, a collaborative-conditional fuzzy clustering method is proposed to granularize the tendency-based feature descriptors and specifically measure the amplitude variation of industrial data which plays a dominant role in the feature space. To quantify the performance of the proposed method, a series of real-world industrial data coming from the energy data center of a steel plant is employed to conduct the comparative experiments. The experimental results demonstrate that the proposed method successively satisfies the requirements of the practically viable prediction.

Integrated energy system (IES) involving coal, by-product gas, power, etc., plays a pivotal role in the steel industry, which greatly requires a reasonable utilization scheme for avoiding extensive energy consumption and intensive environmental pollution. While due to the complex coupling relationship among the multiple energy and the contradiction that existed between the optimization of global IES and local subsystem, an effective solution for this important practical problem is still highly required in the real-world application. As such, in this article, a multicriteria evaluation and cascaded optimization framework is proposed. In order to make a comprehensive evaluation, criteria, including global energy consumption and local gas emission, are, respectively, evaluated from the perspective of both the entire IES and the by-product gas subsystem. Then, a cascaded optimization model is established accordingly, where a reinforcement learning based cascaded algorithm is also designed for efficiently and effectively acquiring an optimal tradeoff as the final solution. The following experiments carried out on practical industrial data clearly demonstrated that the proposed framework gives a scientifically reasonable as well as practically available energy utilization scheme, which is beneficial for energy-saving and emission reduction in practical application.

This article is concerned with the outlier-resistant state estimation problem for industrial pipe networks (PNs). PN signals, e.g., pressure and temperature, are modeled as low-pass time-varying graph signals, and an outlier-resistant graph-frequency domain (GFD) filter is proposed. This article extends the innovation saturation (IS) mechanism from the node domain to the GFD. A GFD IS function is developed with a more compact structure, which reduces outlier effects by restricting PN signals' smoothness. Since the PN signal is time-varying, the saturation function is embedded in the Kalman filter as an inequality constraint. The close-form solution to the filter is derived by replacing the nonlinear inequality constraints with high-pass graph Fourier transforms, and the cutoff frequency is found by analyzing the graph-frequency component of prior and posterior. A steel industrial PN is used to evaluate the proposal. Results indicate that the proposed filter exhibits superior performance in PN systems, particularly internal nodes.